Security context
What an agent needs to avoid regressing past fixes and find the next vuln in this repo.
run-llama/llama_index
main @ 520aa4e
28
Fixes
25
CVEs
CRITICAL
Peak severity
Highlights
SQL Injection: 7 prior fixes. Scrutinize any change in this area.
llama-index-cli: most-fixed (2 issues). Treat as high-risk during review.
24 high-severity fixes in this history; regressions here are high-impact.
Recurring patterns
The bug types that recur here, drawn from past fixes, not open vulnerabilities.
Remote Code Execution: Insecure evaluation of LLM-generated code via eval and exec repeatedly permitted sandbox escapes. Early protections targeted basic string rules (such as blocking underscores) but were bypassed until rigid AST verification and dunder attribute checks were introduced to restrict runtime environments.
SQL Injection: Database operations across various vector stores (e.g., PostgreSQL, DuckDB, ClickHouse) relied on raw string interpolation instead of parameterized queries or ORM query builders, allowing malicious inputs to hijack database execution.
Insecure Deserialization: Unrestricted deserialization of untrusted files using pickle in core components can lead to arbitrary code execution when loading saved indexes or object maps.