Community update, 28 October 2026. Source status: open proposal for review. The approved Founding Charter v1.1 is unchanged.
A proposed policy sets out how Comutability can use AI agents for software work without handing them final authority over releases.
Why this direction matters
We want AI to help turn useful ideas into code, tests and documentation. That speed must not become an excuse for unchecked access to secrets, premature security claims or unreviewed production deployments. The draft introduces permission tiers, risk-based review and evidence-backed release checklists.
What the proposal contains
These are the substantive details from the public GitHub proposal. Some implementation instructions reflect the circumstances when it was first opened and do not, on their own, confirm approval or deployment.
Purpose
Propose Comutability's first community-wide, human-accountable AI development and release policy as Founding Step 5.
This is a draft for review, not an adopted policy. The Founding Charter v1.1 remains unchanged.
Included
policies/AI-DEVELOPMENT-AND-RELEASE.md: AI roles, agent permission tiers A0, A3, privacy, risk classification, prompt-injection defences, human review, build and release gates, incident handling and process for exceptions.policies/RELEASE-CHECKLIST.md: checklist with evidence links and named human release approval.
Key decisions proposed
- Human authority: AI may propose code, tests and release candidates, but cannot independently merge protected changes or approve/sign/publish production releases.
- Permission minimisation: Scoped, time-limited tool access, sandboxed execution and no standing autonomous access to secrets or production.
- Risk-based review: Independent qualified human review of high-risk features; if unavailable, defer that feature's release.
- Honest release claims: No invented test passes or unverified claims of security certification, auditing, provenance levels or accessibility.
- FileGuide v0.1: Read-only scanning and export only. Destructive features require a separate proposal and higher-risk assessment.
Review questions
- Is the A0, A3 tiering practical for a volunteer community?
- Is the high-risk independent-review requirement workable, or should specific categories be expanded or narrowed?
- How should we obtain explicit consent before giving an AI agent external access to non-public project information?
- What evidence should be mandatory for all first releases, versus aspirational improvements?
- How will community members report violations or request exceptions under the separate governance draft PR #1?
Adoption gate
Do not merge as an approved policy until the founder has reviewed the draft, reviewer feedback has been considered, the privacy/security reporting route is adequately defined, and a numbered decision recording approval is added. Draft only.
Background references
OWASP LLM Top 10 · OpenSSF Best Practices · SLSA 1.2 · NIST AI RMF.
What has been decided, and what has not
Comutability's founding charter is approved. It establishes openness, accessibility, shared knowledge, privacy, community participation and human accountability for AI. This proposal is a separate document. It is still open for review and is not an adopted policy.
Next steps
The community still needs to decide which review gates are mandatory, how to obtain independent expertise and how to give agents only the permissions they require.
