Law Outranks Intelligence: Why CORE's Constitutional Design Blocks AI (and Humans) from Bypassing Rules

Law Outranks Intelligence. Including Ours.

Dariusz Newecki designed CORE, an autonomous governance system for code repos. Its key constraint: the daemon cannot write to the .intent/ directory—blocked at analysis, runtime, and commit time by three independent stages.

Why This Matters

The technical reality is that LLMs can reason about codebases better than static rules, creating a temptation to let intelligence adjudicate exceptions. This trap makes systems lawless—every ‘justified exception’ becomes indistinguishable from a violation in the audit trail. Without hard constraints, regulated environments lose the ability to answer what permitted a change, not just what changed. CORE inverts this by making trustworthiness a property of the system architecture, not any actor within it.

Key Insights

• .intent/ directory holds rules as YAML data checked by deterministic engines with a hard halt—no advisory suggestions.
• Write path to .intent/ is blocked at three independent stages: analysis time, runtime, and commit time.
• Even the governor (root user/repo admin) cannot bypass rules; Newecki was flagged for violating docs-integrity rules and spent Saturday fixing his own repo.
• ‘Intelligence proposes. Law disposes.’ The LLM-as-judge runs in reporting mode until promoted by governed decision—authority granted only by constitution.

Practical Applications

•   - Use case: Regulated code repositories needing immutable audit trails.
    - Pitfall: Allowing any actor (including humans) to create ad-hoc exceptions; destroys audit authority.

•   - Use case: AI coding agents operating under compliance frameworks.
    - Pitfall: Letting LLMs override linter/architecture rules 'just this once'; creates invisible violations.

•   - Use case: Multi-developer open-source projects with diverse contributors.
    - Pitfall: Trusting intelligence over system-enforced law leads to inconsistent enforcement across contributors.

•   - Use case: Autonomous daemons governing repo health without human oversight.
    - Pitfall: No hard block on rule modification allows silent rule changes without review or audit trail entry.

References: