If you discover a security vulnerability in Maxim, please report it responsibly:

1. Do NOT open a public issue for security vulnerabilities
2. Open a private security advisory at https://github.com/dennys246/Maxim/security/advisories/new
3. Include: description, reproduction steps, potential impact, and suggested fix if any
4. Expected response time: 48 hours for acknowledgment, 7 days for initial assessment

Maxim includes several safety-critical systems:

• FearAgent — pre-execution safety review of all tool calls
• PainDetector — detects aversive patterns and harmful actions
• HarmRegistry — zero-latency prediction of harmful outcomes
• AutonomyController — gates agent actions by autonomy level

• Atomic file writes — all persistence uses fsync + tmp + replace pattern
• Internet access policy — configurable whitelist/blacklist for web access
• Filesystem policy — sandboxed file access with configurable boundaries
• Cloud redaction — PII filtering before cloud LLM dispatch

Not yet implemented. The features below are design goals for the Mother Maxim / Oasis layer (targeted for 1.1+). The only current privacy mechanism for substrate sharing is a heuristic identity-bearing label filter in hivemind/identity.py (used to quarantine identity-shaped NAc links and EC nodes from substrate bundles). The full ATL+SEM deidentification pipeline described below does not exist in 1.0.

• Bio-system-aware deidentification (planned) — ATL + SEM identity map extracts names/locations deterministically
• Dual-pass pipeline (planned) — client-side deidentification + server-side verification
• Model tier gate (planned) — contributions declare deidentification model; weak models rejected

• Set appropriate autonomy level (--autonomy planning|supervised|autonomous)
• Use MAXIM_LLM_REDACTION_POLICY=strict for cloud providers
• Review maxim doctor output before exposing to network
• For Mother Maxim: run security stress campaign before going public