Kalairos positions itself as a memory layer that resists poisoning, preserves provenance, and surfaces contradictions rather than silently overwriting them. Security reports are taken seriously and triaged ahead of feature work.

Please do not file public GitHub issues for suspected vulnerabilities.

Email: security@krishnalabs.ai

Include, where possible:

• A clear description of the issue and the impact you observed
• Reproduction steps or a minimal proof-of-concept
• The Kalairos version (npm ls kalairos) and Node.js version (node -v)
• Whether you'd like credit in the advisory

You should expect:

• An acknowledgment within 3 business days
• An initial assessment (severity + planned timeline) within 10 business days
• A fix or mitigation in a patch release; the advisory is published once users have had a reasonable window to upgrade

• Memory poisoning and prompt-injection attacks against the ingest path
• Trust-score or provenance bypasses (e.g., source spoofing)
• Authentication or workspace-isolation bypasses in server.js / auth.js
• Data corruption or silent loss in the JSONL store
• Denial-of-service against the rate limiter, write queue, or worker pool
• Supply-chain concerns affecting the published npm package

• Attacks requiring local filesystem access to data.kalairos (the file-store assumes a single trusted process — see CLAUDE.md §13)
• Issues in third-party dependencies that are already publicly tracked
• Theoretical attacks against the deterministic mock embedder used in tests and benchmarks (it is not intended for production)

We follow a 90-day disclosure window from initial report to public advisory, extendable by mutual agreement when a fix needs more runway. Researchers acting in good faith will not be pursued under DMCA, CFAA, or equivalent laws.