Archive
Format: Security brief
-
REVIEW
Big Sleep’s SQLite discovery concerns the database, not the AI research tool
CVE-2025-6965 affected SQLite versions before 3.50.2; Big Sleep was involved in finding the issue.
-
McHire disclosure shows why AI services still need basic access controls
Researchers reported weak administrative credentials and object-level access failures in the hiring platform.
-
REVIEW
Historical advisory: MCP Inspector proxy authentication flaw enabled code execution
The affected tool was MCP Inspector, not every application using the Model Context Protocol.
-
REVIEW
Historical advisory: LiteLLM callback configuration could execute commands
CVE-2024-6825 concerns a code-capable callback configuration path, not a product named “Remote Code Execution”.
-
REVIEW
Historical advisory: prompt injection could manipulate LangChain graph queries
GraphCypherQAChain connects model output to database operations; permissions matter alongside dependency updates.
-
REVIEW
Historical advisory: LangChain LocalFileStore paths could escape their directory
The LocalFileStore issue concerns filesystem boundaries around user-influenced paths.
-
REVIEW
Historical advisory: untrusted Transformers checkpoints could execute code
A checkpoint-loading issue illustrates why model artefacts must be treated as untrusted inputs.
-
REVIEW
Historical advisory: LangChain configuration paths could escape the intended repository
The affected component is LangChain configuration loading, not GitHub itself.
-
REVIEW
Historical advisory: LangChain prompt injection could trigger server-side requests
CVE-2023-32786 concerns attacker-influenced URL retrieval in older LangChain releases.
-
REVIEW
Historical advisory: prompt injection could reach LangChain SQL services
A separate 2023 LangChain advisory links prompt injection to execution against a connected SQL service.
-
REVIEW
Historical advisory: LangChain SQLDatabaseChain could expose database information
CVE-2023-36189 concerns SQL injection through SQLDatabaseChain. The model’s output should not become unrestricted database authority.
-
REVIEW
Historical advisory: LangChain prompt loading could execute untrusted code
A 2023 advisory concerns LangChain prompt loading. Review legacy deployments rather than treating it as a new incident.