Vulnerability database
Every flaw the U.S. government’s cybersecurity agency (CISA) has confirmed attackers are using, plus the new WordPress plugin and theme bugs published each day, sorted by what they mean for a small business. We refresh it twice a day. If you see something your website or office runs, tell us.
457 results · page 5 of 10
Apr 9, 2026PraisonAIAgents is a multi-agent teams system.CVE-2026-40117 · Other software · MervinPraison · reported by GitHubMedium 6.2
PraisonAIAgents is a multi-agent teams system. Prior to 1.5.128, read_skill_file() in skill_tools.py allows reading arbitrary files from the filesystem by accepting an unrestricted skill_path parameter. Unlike file_tools.read_file which enforces workspace boundary confinement, and unlike run_skill_script which requires critical-level approval, read_skill_file has neither protection. An agent influenced by prompt injection can exfiltrate sensitive files without triggering any approval prompt. This vulnerability is fixed in 1.5.128.
- Product
- PraisonAIAgents
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Apr 9, 2026PraisonAIAgents is a multi-agent teams system.CVE-2026-40111 · Other software · MervinPraison · reported by GitHubCritical 9.3
PraisonAIAgents is a multi-agent teams system. Prior to 1.5.128, he memory hooks executor in praisonaiagents passes a user-controlled command string directly to subprocess.run() with shell=True at src/praisonai-agents/praisonaiagents/memory/hooks.py. No sanitization is performed and shell metacharacters are interpreted by /bin/sh before the intended command executes. Two independent attack surfaces exist. The first is via pre_run_command and post_run_command hook event types registered through the hooks configuration. The second and more severe surface is the .praisonai/hooks.json lifecycle configuration, where hooks registered for events such as BEFORE_TOOL and AFTER_TOOL fire automatically during agent operation. An agent that gains file-write access through prompt injection can overwrite .praisonai/hooks.json and have its payload execute silently at every subsequent lifecycle event without further user interaction. This vulnerability is fixed in 1.5.128.
- Product
- PraisonAIAgents
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Apr 9, 2026OpenClaw before 2026.3.24 contains a symlink traversal vulnerability in agents.create and agents.update handlers that use fs.appendFile on IDENTITY.md without sCVE-2026-35632 · Other software · OpenClaw · reported by VulnCheckMedium 6.9
OpenClaw before 2026.3.24 contains a symlink traversal vulnerability in agents.create and agents.update handlers that use fs.appendFile on IDENTITY.md without symlink containment checks. Attackers with workspace access can plant symlinks to append attacker-controlled content to arbitrary files, enabling remote code execution via crontab injection or unauthorized access via SSH key manipulation.
- Product
- OpenClaw
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Full record and what to do →NVD recordgithub.comvulncheck.com
Apr 9, 2026A vulnerability was determined in FoundationAgents MetaGPT up to 0.8.1.CVE-2026-5974 · Other software · FoundationAgents · reported by VulDBMedium 6.9
A vulnerability was determined in FoundationAgents MetaGPT up to 0.8.1. The affected element is the function Bash.run in the library metagpt/tools/libs/terminal.py. This manipulation causes os command injection. The attack is possible to be carried out remotely. The project was informed of the problem early through a pull request but has not reacted yet.
- Product
- MetaGPT
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Apr 9, 2026A vulnerability was found in FoundationAgents MetaGPT up to 0.8.1.CVE-2026-5973 · Other software · FoundationAgents · reported by VulDBMedium 5.5
A vulnerability was found in FoundationAgents MetaGPT up to 0.8.1. Impacted is the function get_mime_type of the file metagpt/utils/common.py. The manipulation results in os command injection. The attack can be executed remotely. The exploit has been made public and could be used. The project was informed of the problem early through a pull request but has not reacted yet.
- Product
- MetaGPT
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Apr 9, 2026A vulnerability has been found in FoundationAgents MetaGPT up to 0.8.1.CVE-2026-5972 · Other software · FoundationAgents · reported by VulDBMedium 5.5
A vulnerability has been found in FoundationAgents MetaGPT up to 0.8.1. This issue affects the function Terminal.run_command in the library metagpt/tools/libs/terminal.py. The manipulation leads to os command injection. Remote exploitation of the attack is possible. The exploit has been disclosed to the public and may be used. The identifier of the patch is d04ffc8dc67903e8b327f78ec121df5e190ffc7b. Applying a patch is the recommended action to fix this issue.
- Product
- MetaGPT
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Apr 9, 2026LangChain is a framework for building agents and LLM-powered applications.CVE-2026-40087 · Other software · langchain-ai · reported by GitHubMedium 5.3
LangChain is a framework for building agents and LLM-powered applications. Prior to 0.3.84 and 1.2.28, LangChain's f-string prompt-template validation was incomplete in two respects. First, some prompt template classes accepted f-string templates and formatted them without enforcing the same attribute-access validation as PromptTemplate. In particular, DictPromptTemplate and ImagePromptTemplate could accept templates containing attribute access or indexing expressions and subsequently evaluate those expressions during formatting. Second, f-string validation based on parsed top-level field names did not reject nested replacement fields inside format specifiers. In this pattern, the nested replacement field appears in the format specifier rather than in the top-level field name. As a result, earlier validation based on parsed field names did not reject the template even though Python formatting would still attempt to resolve the nested expression at runtime. This vulnerability is fixed in 0.3.84 and 1.2.28.
- Product
- langchain
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Apr 9, 2026A flaw has been found in FoundationAgents MetaGPT up to 0.8.1.CVE-2026-5971 · Other software · FoundationAgents · reported by VulDBMedium 5.5
A flaw has been found in FoundationAgents MetaGPT up to 0.8.1. This vulnerability affects the function ActionNode.xml_fill of the file metagpt/actions/action_node.py of the component XML Handler. Executing a manipulation can lead to improper neutralization of directives in dynamically evaluated code. The attack may be launched remotely. The exploit has been published and may be used. The project was informed of the problem early through a pull request but has not reacted yet.
- Product
- MetaGPT
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Apr 9, 2026A vulnerability was detected in FoundationAgents MetaGPT up to 0.8.1.CVE-2026-5970 · Other software · FoundationAgents · reported by VulDBMedium 5.5
A vulnerability was detected in FoundationAgents MetaGPT up to 0.8.1. This affects the function check_solution of the component HumanEvalBenchmark/MBPPBenchmark. Performing a manipulation results in code injection. The attack may be initiated remotely. The exploit is now public and may be used. The project was informed of the problem early through a pull request but has not reacted yet.
- Product
- MetaGPT
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Apr 8, 2026PraisonAI is a multi-agent teams system.CVE-2026-39888 · Other software · MervinPraison · reported by GitHubCritical 9.9
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.
- Product
- praisonaiagents
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Apr 7, 2026An issue that allowed MCP agents to access certificate information from outside of their authorized organization scope has been resolved.CVE-2026-5379 · Other software · runZero · reported by 44488dab-36db-4358-99f9-bc116477f914Low 3
An issue that allowed MCP agents to access certificate information from outside of their authorized organization scope has been resolved. This is an instance of CWE-863: Incorrect Authorization, and has an estimated CVSS score of CVSS:3.1/AV:N/AC:H/PR:H/UI:N/S:C/C:L/I:N/A:N (3.0 Low). This issue was fixed in version 4.0.260203.0 of the runZero Platform.
- Product
- Platform
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Full record and what to do →NVD recordhelp.runzero.comrunzero.com
Apr 7, 2026An issue that allowed MCP agents to access remediation and asset information from outside of the authorized organization scope has been resolved.CVE-2026-5374 · Other software · runZero · reported by 44488dab-36db-4358-99f9-bc116477f914Medium 5.8
An issue that allowed MCP agents to access remediation and asset information from outside of the authorized organization scope has been resolved. This is an instance of CWE-863: Incorrect Authorization, and has an estimated CVSS score of CVSS:3.1/AV:N/AC:H/PR:H/UI:N/S:C/C:H/I:N/A:N (5.8 Medium). This issue was fixed in version 4.0.260202.0 of the runZero Platform.
- Product
- Platform
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Full record and what to do →NVD recordhelp.runzero.comrunzero.com
Mar 30, 2026LangChain is a framework for building agents and LLM-powered applications.CVE-2026-34070 · Other software · langchain-ai · reported by GitHubHigh 7.5
LangChain is a framework for building agents and LLM-powered applications. Prior to version 1.2.22, multiple functions in langchain_core.prompts.loading read files from paths embedded in deserialized config dicts without validating against directory traversal or absolute path injection. When an application passes user-influenced prompt configurations to load_prompt() or load_prompt_from_config(), an attacker can read arbitrary files on the host filesystem, constrained only by file-extension checks (.txt for templates, .json/.yaml for examples). This issue has been patched in version 1.2.22.
- Product
- langchain
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Mar 29, 2026OpenClaw before 2026.3.11 contains a session sandbox escape vulnerability in the session_status tool that allows sandboxed subagents to access parent or siblingCVE-2026-32918 · Other software · OpenClaw · reported by VulnCheckCritical 9.2
OpenClaw before 2026.3.11 contains a session sandbox escape vulnerability in the session_status tool that allows sandboxed subagents to access parent or sibling session state. Attackers can supply arbitrary sessionKey values to read or modify session data outside their sandbox scope, including persisted model overrides.
- Product
- OpenClaw
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Full record and what to do →NVD recordgithub.comvulncheck.com
Mar 29, 2026OpenClaw before 2026.3.11 contains a sandbox boundary bypass vulnerability allowing leaf subagents to access the subagents control surface and resolve against pCVE-2026-32915 · Other software · OpenClaw · reported by VulnCheckCritical 9.3
OpenClaw before 2026.3.11 contains a sandbox boundary bypass vulnerability allowing leaf subagents to access the subagents control surface and resolve against parent requester scope instead of their own session tree. A low-privilege sandboxed leaf worker can steer or kill sibling runs and cause execution with broader tool policies by exploiting insufficient authorization checks on subagent control requests.
- Product
- OpenClaw
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Full record and what to do →NVD recordgithub.comvulncheck.com
Mar 27, 2026Langflow is a tool for building and deploying AI-powered agents and workflows.CVE-2026-34046 · Other software · langflow-ai · reported by GitHubHigh 8.7
Langflow is a tool for building and deploying AI-powered agents and workflows. Prior to version 1.5.1, the `_read_flow` helper in `src/backend/base/langflow/api/v1/flows.py` branched on the `AUTO_LOGIN` setting to decide whether to filter by `user_id`. When `AUTO_LOGIN` was `False` (i.e., authentication was enabled), neither branch enforced an ownership check — the query returned any flow matching the given UUID regardless of who owned it. This allowed any authenticated user to read any other user's flow, including embedded plaintext API keys; modify the logic of another user's AI agents, and/or delete flows belonging to other users. The vulnerability was introduced by the conditional logic that was meant to accommodate public/example flows (those with `user_id = NULL`) under auto-login mode, but inadvertently left the authenticated path without an ownership filter. The fix in version 1.5.1 removes the `AUTO_LOGIN` conditional entirely and unconditionally scopes the query to the requesting user.
- Product
- langflow
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Mar 27, 2026Langflow is a tool for building and deploying AI-powered agents and workflows.CVE-2026-33873 · Other software · langflow-ai · reported by GitHubCritical 9.3
Langflow is a tool for building and deploying AI-powered agents and workflows. Prior to version 1.9.0, the Agentic Assistant feature in Langflow executes LLM-generated Python code during its validation phase. Although this phase appears intended to validate generated component code, the implementation reaches dynamic execution sinks and instantiates the generated class server-side. In deployments where an attacker can access the Agentic Assistant feature and influence the model output, this can result in arbitrary server-side Python execution. Version 1.9.0 fixes the issue.
- Product
- langflow
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Mar 27, 2026A weakness has been identified in huggingface smolagents 1.25.0.dev0.CVE-2026-4963 · Other software · huggingface · reported by VulDBLow 2.1
A weakness has been identified in huggingface smolagents 1.25.0.dev0. This affects the function evaluate_augassign/evaluate_call/evaluate_with of the file src/smolagents/local_python_executor.py of the component Incomplete Fix CVE-2025-9959. This manipulation causes code injection. It is possible to initiate the attack remotely. The exploit has been made available to the public and could be used for attacks. The vendor was contacted early about this disclosure but did not respond in any way.
- Product
- smolagents
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Full record and what to do →NVD recordgist.github.comgist.github.com
Mar 26, 2026PinchTab is a standalone HTTP server that gives AI agents direct control over a Chrome browser.CVE-2026-33623 · Other software · pinchtab · reported by GitHubMedium 6.7
PinchTab is a standalone HTTP server that gives AI agents direct control over a Chrome browser. PinchTab `v0.8.4` contains a Windows-only command injection issue in the orphaned Chrome cleanup path. When an instance is stopped, the Windows cleanup routine builds a PowerShell `-Command` string using a `needle` derived from the profile path. In `v0.8.4`, that string interpolation escapes backslashes but does not safely neutralize other PowerShell metacharacters. If an attacker can launch an instance using a crafted profile name and then trigger the cleanup path, they may be able to execute arbitrary PowerShell commands on the Windows host in the security context of the PinchTab process user. This is not an unauthenticated internet RCE. It requires authenticated, administrative-equivalent API access to instance lifecycle endpoints, and the resulting command execution inherits the permissions of the PinchTab OS user rather than bypassing host privilege boundaries. Version 0.8.5 contains a patch for the issue.
- Product
- pinchtab
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Mar 26, 2026PinchTab is a standalone HTTP server that gives AI agents direct control over a Chrome browser.CVE-2026-33622 · Other software · pinchtab · reported by GitHubMedium 6.1
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.
- Product
- pinchtab
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Mar 26, 2026PinchTab is a standalone HTTP server that gives AI agents direct control over a Chrome browser.CVE-2026-33621 · Other software · pinchtab · reported by GitHubMedium 4.8
PinchTab is a standalone HTTP server that gives AI agents direct control over a Chrome browser. PinchTab `v0.7.7` through `v0.8.4` contain incomplete request-throttling protections for auth-checkable endpoints. In `v0.7.7` through `v0.8.3`, a fully implemented `RateLimitMiddleware` existed in `internal/handlers/middleware.go` but was not inserted into the production HTTP handler chain, so requests were not subject to the intended per-IP throttle. In the same pre-`v0.8.4` range, the original limiter also keyed clients using `X-Forwarded-For`, which would have allowed client-controlled header spoofing if the middleware had been enabled. `v0.8.4` addressed those two issues by wiring the limiter into the live handler chain and switching the key to the immediate peer IP, but it still exempted `/health` and `/metrics` from rate limiting even though `/health` remained an auth-checkable endpoint when a token was configured. This issue weakens defense in depth for deployments where an attacker can reach the API, especially if a weak human-chosen token is used. It is not a direct authentication bypass or token disclosure issue by itself. PinchTab is documented as local-first by default and u
- Product
- pinchtab
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Mar 26, 2026PinchTab is a standalone HTTP server that gives AI agents direct control over a Chrome browser.CVE-2026-33620 · Other software · pinchtab · reported by GitHubMedium 4.3
PinchTab is a standalone HTTP server that gives AI agents direct control over a Chrome browser. PinchTab `v0.7.8` through `v0.8.3` accepted the API token from a `token` URL query parameter in addition to the `Authorization` header. When a valid API credential is sent in the URL, it can be exposed through request URIs recorded by intermediaries or client-side tooling, such as reverse proxy access logs, browser history, shell history, clipboard history, and tracing systems that capture full URLs. This issue is an unsafe credential transport pattern rather than a direct authentication bypass. It only affects deployments where a token is configured and a client actually uses the query-parameter form. PinchTab's security guidance already recommended `Authorization: Bearer <token>`, but `v0.8.3` still accepted `?token=` and included first-party flows that generated and consumed URLs containing the token. This was addressed in v0.8.4 by removing query-string token authentication and requiring safer header- or session-based authentication flows.
- Product
- pinchtab
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Mar 26, 2026PinchTab is a standalone HTTP server that gives AI agents direct control over a Chrome browser.CVE-2026-33619 · Other software · pinchtab · reported by GitHubMedium 4.1
PinchTab is a standalone HTTP server that gives AI agents direct control over a Chrome browser. PinchTab v0.8.3 contains a server-side request forgery issue in the optional scheduler's webhook delivery path. When a task is submitted to `POST /tasks` with a user-controlled `callbackUrl`, the v0.8.3 scheduler sends an outbound HTTP `POST` to that URL when the task reaches a terminal state. In that release, the webhook path validated only the URL scheme and did not reject loopback, private, link-local, or other non-public destinations. Because the v0.8.3 implementation also used the default HTTP client behavior, redirects were followed and the destination was not pinned to validated IPs. This allowed blind SSRF from the PinchTab server to attacker-chosen HTTP(S) targets reachable from the server. This issue is narrower than a general unauthenticated internet-facing SSRF. The scheduler is optional and off by default, and in token-protected deployments the attacker must already be able to submit tasks using the server's master API token. In PinchTab's intended deployment model, that token represents administrative control rather than a low-privilege role. Tokenless deployments lower the
- Product
- pinchtab
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Mar 24, 2026Langflow is a tool for building and deploying AI-powered agents and workflows.CVE-2026-33497 · Other software · langflow-ai · reported by GitHubHigh 8.7
Langflow is a tool for building and deploying AI-powered agents and workflows. Prior to version 1.7.1, in the download_profile_picture function of the /profile_pictures/{folder_name}/{file_name} endpoint, the folder_name and file_name parameters are not strictly filtered, which allows the secret_key to be read across directories. Version 1.7.1 contains a patch.
- Product
- langflow
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Mar 24, 2026Langflow is a tool for building and deploying AI-powered agents and workflows.CVE-2026-33484 · Other software · langflow-ai · reported by GitHubHigh 7.5
Langflow is a tool for building and deploying AI-powered agents and workflows. In versions 1.0.0 through 1.8.1, the `/api/v1/files/images/{flow_id}/{file_name}` endpoint serves image files without any authentication or ownership check. Any unauthenticated request with a known flow_id and file_name returns the image with HTTP 200. In a multi-tenant deployment, any attacker who can discover or guess a `flow_id` (UUIDs can be leaked through other API responses) can download any user's uploaded images without credentials. Version 1.9.0 contains a patch.
- Product
- langflow
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Mar 24, 2026Langflow is a tool for building and deploying AI-powered agents and workflows.CVE-2026-33475 · Other software · langflow-ai · reported by GitHubCritical 9.1
Langflow is a tool for building and deploying AI-powered agents and workflows. An unauthenticated remote shell injection vulnerability exists in multiple GitHub Actions workflows in the Langflow repository prior to version 1.9.0. Unsanitized interpolation of GitHub context variables (e.g., `${{ github.head_ref }}`) in `run:` steps allows attackers to inject and execute arbitrary shell commands via a malicious branch name or pull request title. This can lead to secret exfiltration (e.g., `GITHUB_TOKEN`), infrastructure manipulation, or supply chain compromise during CI/CD execution. Version 1.9.0 patches the vulnerability. --- ### Details Several workflows in `.github/workflows/` and `.github/actions/` reference GitHub context variables directly in `run:` shell commands, such as: ```yaml run: | validate_branch_name "${{ github.event.pull_request.head.ref }}" ``` Or: ```yaml run: npx playwright install ${{ inputs.browsers }} --with-deps ``` Since `github.head_ref`, `github.event.pull_request.title`, and custom `inputs.*` may contain **user-controlled values**, they must be treated as **untrusted input**. Direct interpolation without proper quoting or sanitization leads to sh
- Product
- langflow
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Mar 24, 2026Langflow is a tool for building and deploying AI-powered agents and workflows.CVE-2026-33309 · Other software · langflow-ai · reported by GitHubCritical 9.9
Langflow is a tool for building and deploying AI-powered agents and workflows. Versions 1.2.0 through 1.8.1 have a bypass of the patch for CVE-2025-68478 (External Control of File Name), leading to the root architectural issue within `LocalStorageService` remaining unresolved. Because the underlying storage layer lacks boundary containment checks, the system relies entirely on the HTTP-layer `ValidatedFileName` dependency. This defense-in-depth failure leaves the `POST /api/v2/files/` endpoint vulnerable to Arbitrary File Write. The multipart upload filename bypasses the path-parameter guard, allowing authenticated attackers to write files anywhere on the host system, leading to Remote Code Execution (RCE). Version 1.9.0 contains an updated fix.
- Product
- langflow
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Mar 21, 2026A vulnerability was found in Foundation Agents MetaGPT up to 0.8.1.CVE-2026-4516 · Other software · Foundation Agents · reported by VulDBLow 2.1
A vulnerability was found in Foundation Agents MetaGPT up to 0.8.1. This vulnerability affects unknown code of the file metagpt/actions/di/write_analysis_code.py of the component DataInterpreter. The manipulation results in injection. It is possible to launch the attack remotely. The exploit has been made public and could be used. The vendor was contacted early about this disclosure but did not respond in any way.
- Product
- MetaGPT
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Mar 21, 2026A vulnerability has been found in Foundation Agents MetaGPT up to 0.8.1.CVE-2026-4515 · Other software · Foundation Agents · reported by VulDBLow 2.1
A vulnerability has been found in Foundation Agents MetaGPT up to 0.8.1. This affects the function code_generate of the file metagpt/ext/aflow/scripts/operator.py. The manipulation leads to code injection. It is possible to initiate the attack remotely. The exploit has been disclosed to the public and may be used. The vendor was contacted early about this disclosure but did not respond in any way.
- Product
- MetaGPT
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Mar 20, 2026PinchTab is a standalone HTTP server that gives AI agents direct control over a Chrome browser.CVE-2026-33081 · Other software · pinchtab · reported by GitHubMedium 5.8
PinchTab is a standalone HTTP server that gives AI agents direct control over a Chrome browser. Versions 0.8.2 and below have a Blind SSRF vulnerability in the /download endpoint. The validateDownloadURL() function only checks the initial user-supplied URL, but the embedded Chromium browser can follow attacker-controlled redirects/navigations to internal network addresses after validation. Exploitation requires security.allowDownload=true (disabled by default), limiting real-world impact. An attacker-controlled page can use JavaScript redirects or resource requests to make the browser reach internal services from the PinchTab host, resulting in a blind Server-Side Request Forgery (SSRF) condition against internal-only services. The issue has been patched in version 0.8.3.
- Product
- pinchtab
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Mar 20, 2026Langflow is a tool for building and deploying AI-powered agents and workflows.CVE-2026-33053 · Other software · langflow-ai · reported by GitHubMedium 6.1
Langflow is a tool for building and deploying AI-powered agents and workflows. In versions prior to 1.9.0, the delete_api_key_route() endpoint accepts an api_key_id path parameter and deletes it with only a generic authentication check (get_current_active_user dependency). However, the delete_api_key() CRUD function does NOT verify that the API key belongs to the current user before deletion.
- Product
- langflow
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Mar 19, 2026OpenClaw versions prior to 2026.2.25 contain a symlink traversal vulnerability in the agents.files.get and agents.files.set methods that allows reading and writCVE-2026-32013 · Other software · OpenClaw · reported by VulnCheckHigh 8.7
OpenClaw versions prior to 2026.2.25 contain a symlink traversal vulnerability in the agents.files.get and agents.files.set methods that allows reading and writing files outside the agent workspace. Attackers can exploit symlinked allowlisted files to access arbitrary host files within gateway process permissions, potentially enabling code execution through file overwrite attacks.
- Product
- OpenClaw
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Mar 17, 2026The GLPI Inventory Plugin handles network discovery, inventory, software deployment, and data collection for GLPI agents.CVE-2026-26001 · Other software · glpi-project · reported by GitHubHigh 7.1
The GLPI Inventory Plugin handles network discovery, inventory, software deployment, and data collection for GLPI agents. Prior to 1.6.6, non sanitized user input can lend to an SQL injection from reports, with adequate rights. This vulnerability is fixed in 1.6.6.
- Product
- glpi-inventory-plugin
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Mar 12, 2026Graphiti is a framework for building and querying temporal context graphs for AI agents.CVE-2026-32247 · Other software · getzep · reported by GitHubHigh 8.1
Graphiti is a framework for building and querying temporal context graphs for AI agents. Graphiti versions before 0.28.2 contained a Cypher injection vulnerability in shared search-filter construction for non-Kuzu backends. Attacker-controlled label values supplied through SearchFilters.node_labels were concatenated directly into Cypher label expressions without validation. In MCP deployments, this was exploitable not only through direct untrusted access to the Graphiti MCP server, but also through prompt injection against an LLM client that could be induced to call search_nodes with attacker-controlled entity_types values. The MCP server mapped entity_types to SearchFilters.node_labels, which then reached the vulnerable Cypher construction path. Affected backends included Neo4j, FalkorDB, and Neptune. Kuzu was not affected by the label-injection issue because it used parameterized label handling rather than string-interpolated Cypher labels. This issue was mitigated in 0.28.2.
- Product
- graphiti
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Mar 11, 2026Use after free in Agents in Google Chrome prior to 146.0.7680.71 allowed a remote attacker to potentially exploit heap corruption via a crafted HTML page.CVE-2026-3917 · Computers, phones & browsers · Google · reported by chrome-cve-admin@google.comHigh 8.8
Use after free in Agents in Google Chrome prior to 146.0.7680.71 allowed a remote attacker to potentially exploit heap corruption via a crafted HTML page. (Chromium security severity: High)
- Product
- Chrome
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Full record and what to do →NVD recordchromereleases.googleblog.comissues.chromium.org
Mar 10, 2026Coral Server is open collaboration infrastructure that enables communication, coordination, trust and payments for The Internet of Agents.CVE-2026-30970 · Other software · Coral-Protocol · reported by GitHubHigh 8.8
Coral Server is open collaboration infrastructure that enables communication, coordination, trust and payments for The Internet of Agents. Prior to 1.1.0, Coral Server allowed the creation of agent sessions through the /api/v1/sessions endpoint without strong authentication. This endpoint performs resource-intensive initialization operations including container spawning and memory context creation. An attacker capable of accessing the endpoint could create sessions or consume system resources without proper authorization. This vulnerability is fixed in 1.1.0.
- Product
- coral-server
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Mar 10, 2026Coral Server is open collaboration infrastructure that enables communication, coordination, trust and payments for The Internet of Agents.CVE-2026-30969 · Other software · Coral-Protocol · reported by GitHubHigh 7.6
Coral Server is open collaboration infrastructure that enables communication, coordination, trust and payments for The Internet of Agents. Prior to 1.1.0, Coral Server did not enforce strong authentication between agents and the server within an active session. This could allow an attacker who obtained or predicted a session identifier to impersonate an agent or join an existing session. This vulnerability is fixed in 1.1.0.
- Product
- coral-server
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Mar 10, 2026Coral Server is open collaboration infrastructure that enables communication, coordination, trust and payments for The Internet of Agents.CVE-2026-30968 · Other software · Coral-Protocol · reported by GitHubHigh 8.6
Coral Server is open collaboration infrastructure that enables communication, coordination, trust and payments for The Internet of Agents. Prior to 1.1.0, the SSE endpoint (/sse/v1/...) in Coral Server did not strongly validate that a connecting agent was a legitimate participant in the session. This could theoretically allow unauthorized message injection or observation. This vulnerability is fixed in 1.1.0.
- Product
- coral-server
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Mar 7, 2026PinchTab is a standalone HTTP server that gives AI agents direct control over a Chrome browser.CVE-2026-30834 · Other software · pinchtab · reported by GitHubHigh 7.5
PinchTab is a standalone HTTP server that gives AI agents direct control over a Chrome browser. Prior to version 0.7.7, a Server-Side Request Forgery (SSRF) vulnerability in the /download endpoint allows any user with API access to induce the PinchTab server to make requests to arbitrary URLs, including internal network services and local system files, and exfiltrate the full response content. This issue has been patched in version 0.7.7.
- Product
- pinchtab
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Mar 3, 2026The GLPI Inventory Plugin handles network discovery, inventory, software deployment, and data collection for GLPI agents.CVE-2026-25590 · Other software · glpi-project · reported by GitHubMedium 4.5
The GLPI Inventory Plugin handles network discovery, inventory, software deployment, and data collection for GLPI agents. Prior to 1.6.6, there is a reflected XSS vulnerability in task jobs. This vulnerability is fixed in 1.6.6.
- Product
- glpi-inventory-plugin
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Feb 25, 2026Langflow is a tool for building and deploying AI-powered agents and workflows.CVE-2026-27966 · Other software · langflow-ai · reported by GitHubCritical 9.8
Langflow is a tool for building and deploying AI-powered agents and workflows. Prior to version 1.8.0, the CSV Agent node in Langflow hardcodes `allow_dangerous_code=True`, which automatically exposes LangChain’s Python REPL tool (`python_repl_ast`). As a result, an attacker can execute arbitrary Python and OS commands on the server via prompt injection, leading to full Remote Code Execution (RCE). Version 1.8.0 fixes the issue.
- Product
- langflow
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Feb 18, 2026A weakness has been identified in huggingface smolagents 1.24.0.CVE-2026-2654 · Other software · huggingface · reported by VulDBLow 2.1
A weakness has been identified in huggingface smolagents 1.24.0. Impacted is the function requests.get/requests.post of the component LocalPythonExecutor. Executing a manipulation can lead to server-side request forgery. It is possible to launch the attack remotely. The exploit has been made available to the public and could be used for attacks. The vendor was contacted early about this disclosure but did not respond in any way.
- Product
- smolagents
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Feb 12, 2026AutoGPT is a platform that allows users to create, deploy, and manage continuous artificial intelligence agents that automate complex workflows.CVE-2026-26020 · Other software · Significant-Gravitas · reported by GitHubCritical 9.4
AutoGPT is a platform that allows users to create, deploy, and manage continuous artificial intelligence agents that automate complex workflows. Prior to 0.6.48, an authenticated user could achieve Remote Code Execution (RCE) on the backend server by embedding a disabled block inside a graph. The BlockInstallationBlock — a development tool capable of writing and importing arbitrary Python code — was marked disabled=True, but graph validation did not enforce this flag. This allowed any authenticated user to bypass the restriction by including the block as a node in a graph, rather than calling the block's execution endpoint directly (which did enforce the flag). This vulnerability is fixed in 0.6.48.
- Product
- AutoGPT
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Feb 11, 2026An improper certificate validation vulnerability in PAN-OS allows users to connect Terminal Server Agents on Windows to PAN-OS using expired certificates even iCVE-2026-0228 · Network & remote access · Palo Alto Networks · reported by psirt@paloaltonetworks.comLow 1.3
An improper certificate validation vulnerability in PAN-OS allows users to connect Terminal Server Agents on Windows to PAN-OS using expired certificates even if the PAN-OS configuration would not normally permit them to do so.
- Product
- Cloud NGFW
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Full record and what to do →NVD recordsecurity.paloaltonetworks.com
Feb 10, 2026LangChain is a framework for building agents and LLM-powered applications.CVE-2026-26013 · Other software · langchain-ai · reported by GitHubLow 3.7
LangChain is a framework for building agents and LLM-powered applications. Prior to 1.2.11, the ChatOpenAI.get_num_tokens_from_messages() method fetches arbitrary image_url values without validation when computing token counts for vision-enabled models. This allows attackers to trigger Server-Side Request Forgery (SSRF) attacks by providing malicious image URLs in user input. This vulnerability is fixed in 1.2.11.
- Product
- langchain
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Feb 10, 2026AutoGPT is a platform that allows users to create, deploy, and manage continuous artificial intelligence agents that automate complex workflows.CVE-2026-26006 · Other software · Significant-Gravitas · reported by GitHubMedium 6.5
AutoGPT is a platform that allows users to create, deploy, and manage continuous artificial intelligence agents that automate complex workflows. The autogpt before 0.6.32 is vulnerable to Regular Expression Denial of Service due to the use of regex at Code Extraction Block. The two Regex are used containing the corresponding dangerous patterns \s+[\s\S]*? and \s+(.*?). They share a common characteristic — the combination of two adjacent quantifiers that can match the same space character (\s). As a result, an attacker can supply a long sequence of space characters to trigger excessive regex backtracking, potentially leading to a Denial of Service (DoS). This vulnerability is fixed in 0.6.32.
- Product
- AutoGPT
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Feb 6, 2026The Agentspace service was affected by a vulnerability that exposed sensitive information due to the use of predictable Google Cloud Storage bucket names.CVE-2026-1727 · Computers, phones & browsers · Google Cloud · reported by f45cbf4e-4146-4068-b7e1-655ffc2c548cCritical 9.1
The Agentspace service was affected by a vulnerability that exposed sensitive information due to the use of predictable Google Cloud Storage bucket names. These names were utilized for error logs and temporary staging during data imports from GCS and Cloud SQL. This predictability allowed an attacker to engage in "bucket squatting" by establishing these buckets before a victim's initial use. All versions after December 12th, 2025 have been updated to protect from this vulnerability. No user action is required for this.
- Product
- Gemini Enterprise (formerly Agentspace)
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Feb 6, 2026Semantic Kernel is an SDK used to build, orchestrate, and deploy AI agents and multi-agent systems.CVE-2026-25592 · Computers, phones & browsers · microsoft · reported by GitHubCritical 9.9
Semantic Kernel is an SDK used to build, orchestrate, and deploy AI agents and multi-agent systems. Prior to 1.71.0, an Arbitrary File Write vulnerability has been identified in Microsoft's Semantic Kernel .NET SDK, specifically within the SessionsPythonPlugin. The problem has been fixed in Microsoft.SemanticKernel.Core version 1.71.0. As a mitigation, users can create a Function Invocation Filter which checks the arguments being passed to any calls to DownloadFileAsync or UploadFileAsync and ensures the provided localFilePath is allow listed.
- Product
- semantic-kernel
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Feb 5, 2026AutoGPT is a platform that allows users to create, deploy, and manage continuous artificial intelligence agents that automate complex workflows.CVE-2025-32393 · Other software · Significant-Gravitas · reported by GitHubHigh 8.7
AutoGPT is a platform that allows users to create, deploy, and manage continuous artificial intelligence agents that automate complex workflows. Prior to autogpt-platform-beta-v0.6.32, there is a DoS vulnerability in ReadRSSFeedBlock. In RSSBlock, feedparser.parser is called to obtain the XML file according to the URL input by the user, parse the XML, and finally obtain the parsed result. However, during the parsing process, there is no limit on the parsing time and the resources that can be allocated for parsing. When a malicious user lets RSSBlock parse a carefully constructed, deep XML, it will cause memory resources to be exhausted, eventually causing DoS. This issue has been patched in autogpt-platform-beta-v0.6.32.
- Product
- AutoGPT
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Feb 4, 2026AutoGPT is a platform that allows users to create, deploy, and manage continuous artificial intelligence agents that automate complex workflows.CVE-2026-22038 · Other software · Significant-Gravitas · reported by GitHubHigh 8.1
AutoGPT is a platform that allows users to create, deploy, and manage continuous artificial intelligence agents that automate complex workflows. Prior to autogpt-platform-beta-v0.6.46, the AutoGPT platform's Stagehand integration blocks log API keys and authentication secrets in plaintext using logger.info() statements. This occurs in three separate block implementations (StagehandObserveBlock, StagehandActBlock, and StagehandExtractBlock) where the code explicitly calls api_key.get_secret_value() and logs the result. This issue has been patched in autogpt-platform-beta-v0.6.46.
- Product
- AutoGPT
- What to do
- Apply the vendor’s update. The NVD record links to the advisory.
Where this comes from: the exploited list is CISA’s Known Exploited Vulnerabilities catalog. Everything else is the whole National Vulnerability Database, where each record is published by the team that found the bug (Wordfence, Patchstack, WPScan, the vendors themselves and others); we show who. Exploit odds are FIRST’s EPSS, updated daily. Titles and categories are ours, so check the linked record before acting on a detail. For a write-up of the big ones, see our advisories. Every vendor, and every WordPress plugin and theme with a recorded flaw, also has its own page. All of it is free to reuse as JSON, CSV or RSS.
Weekly security brief
The flaws that hit local websites, every Friday.
One short email: what got exploited this week, which WordPress plugins broke, and what to check. Unsubscribe in one click.
Tell us what you're building.
A website, your marketing, a security check, an app, or the thing you've been running out of a spreadsheet for three years. A few lines is plenty.
Not ready to talk yet? Run the free website check and see what Google sees.