The Jaegr team takes security seriously. If you discover a vulnerability or security flaw in the policy evaluator, Solana instruction parser, or API boundaries, please report it privately rather than opening a public issue.
Please use GitHub's private vulnerability reporting (repository Security → Advisories) or contact the maintainers directly.
Please include:
- A description of the vulnerability.
- Steps to reproduce or a minimal proof of concept.
- The potential impact on user funds or agent policies.
We commit to acknowledging reports within 48 hours and providing a coordinated remediation timeline.
Prompt injection is an inherent property of current generation Large Language Models. Jaegr does not certify that any AI model is 100% immune to prompt injection. The primary security guarantee in an agentic Solana system must come from the independent runtime authorization policy evaluator that clamps amounts and verifies recipients regardless of what the LLM proposes.
Jaegr never requests, accepts, stores, or logs private keys or mnemonic seed phrases. All actions evaluated by this product are unsigned instruction proposals.
The MVP evaluates Solana System Program transfers and explicitly rejects unallowlisted third-party programs. It does not broadcast transactions to Solana Mainnet.