AIC-agents assumes:

• Systems can fail
• Agents can behave unexpectedly
• Attackers can be intelligent

Security is treated as a continuous property, not a feature.

• Malicious agents
• Compromised execution environments
• Emergent harmful behavior
• Trust manipulation

If you discover a security issue:

• Do not open a public issue
• Contact the maintainers privately
• Provide reproduction steps and potential impact

Responsible disclosure is essential.

• Sandboxed execution
• Trust decay and revocation
• Action rollbacks
• Simulation-before-deployment

Security research is an ongoing effort.

We are particularly interested in:

• Agent containment strategies
• Trust poisoning resistance
• Human override mechanisms
• Failure mode transparency

If you are looking to build uncontrolled, extractive, or opaque AI systems, this project is likely not for you.

If you are willing to build slowly, responsibly, and honestly — welcome.