Legal and charter

Article IX — AI Governance

How CRIS governs the intelligent systems it builds, including those that improve themselves.

Approved by the Board for implementation · 20 November 2026

Approved

Approved by the Board for implementation, 20 November 2026. Counsel is confirming the open points, shown in [brackets]. These bylaws work under Delaware law and our Certificate of Incorporation; where they differ, the law and the certificate prevail.

In one paragraph

Summary, in plain words — not a quote

Under the proposed Article IX, every CRIS system would have to stay faithful to reality and our declared purposes. Before material deployment, its outputs would be tested against independent evidence; without verified correspondence, it would not be deployed where it could affect people, assets or public safety. Self-improving systems would be allowed if each improvement keeps that correspondence. No AI system could use the Board’s powers, elect directors, issue stock, amend our documents or bind CRIS without a human-authorised process.

The text

Article IX — AI Governance

Section 9.1. Foundational Alignment. In keeping with the Fourth Law, every intelligent system the Corporation develops shall be governed by faithfulness to reality and to the Corporation's declared beneficial purposes.

Section 9.2. Correspondence Requirement. A system's outputs, models and self-assessments shall be tested against independent evidence before material deployment and monitored afterward. A system that cannot demonstrate verified correspondence with reality shall not be deployed for uses that could materially affect people, assets or public safety.

Section 9.3. Self-Improving Systems. The Corporation may develop systems that evaluate and improve their own models, software, architecture, methods and performance, provided each improvement preserves verified correspondence and accumulated knowledge (recursion without rupture).

Section 9.4. Controlled Change. Material system modifications are subject to risk assessment, authorization, validation, documentation and monitoring proportionate to risk.

Section 9.5. Corporate Authority. No artificial intelligence system may exercise the statutory powers of the Board, elect or remove directors, issue stock, amend governing documents, or bind the Corporation except through a lawful human-authorized corporate process.

Section 9.6. Human Oversight. The Corporation shall keep appropriate human oversight and intervention mechanisms for systems that could materially affect individuals, communities, corporate assets or public safety.

Section 9.7. Incident Management. Material failures, security incidents and significant system risks shall be reported through procedures the Board sets, and each shall be reviewed for its lessons, which are added to the Corporation's knowledge record.

Section 9.8. AI Governance Constitution. The Board may adopt an AI Governance Constitution (the "CRIS AI Constitution") and related technical policies consistent with these Bylaws and the Fourth Law.

Quoted exactly from the bylaws (version of 8 November 2026, approved by the Board for implementation on 20 November 2026). Headings, lists and tables are only laid out for reading.

In pictures

  1. Change proposedA new system, or a material change to one (§9.4)
  2. TestedOutputs, models and self-assessments checked against independent evidence
  3. Verified correspondence?The gate: can the system show it matches reality?
  • YesDeploy, then keep monitoring (§9.2) with human oversight (§9.6)
  • NoNot deployed for uses that could materially affect people, assets or public safety
The correspondence gate in Section 9.2 of the proposed bylaws. A summary; the Correspondence Prover and C-Gate are the systems that would carry it out.

May, under §9.3

  • Evaluate its own performance
  • Improve its models
  • Improve its software and architecture
  • Improve its methods

Only while each improvement keeps verified correspondence and accumulated knowledge.

May never, under §9.5

  • Exercise the Board’s statutory powers
  • Elect or remove directors
  • Issue stock
  • Amend governing documents
  • Bind CRIS, except through a lawful process people authorise
What an AI system may never do, Section 9.5, beside what Section 9.3 allows. A summary of both sections.

What it means in practice

The proposed Article IX applies the Fourth Law of Logic to the systems we build. Its centre is a gate: before a material deployment, a system would have to show, against independent evidence, that its outputs match reality. If it cannot, it would not be used where people, assets or public safety could be affected.

For engineers and researchers, this would mean every material change carries a test and a record. The Correspondence Prover issues the proof, the C-Gate decides whether a change may pass, and the Continuity Audit Suite re-runs a sample. Changes are assessed in proportion to their risk, and humans keep the means to step in.

Self-improving systems would be allowed, but only on one condition. Each improvement must keep verified correspondence and the knowledge already gained. The bylaws call this recursion without rupture, the idea at the heart of the Founder’s book of the same name.

For the Board and stockholders, the most important line is Section 9.5. No AI system holds corporate authority. A system may inform a decision, but it cannot make one for CRIS. It cannot use the Board’s powers, elect or remove directors, issue stock, amend our documents or sign on its own. Every such act needs a lawful process that people authorise.

When something goes wrong, the incident would be reported, reviewed and its lesson added to our knowledge record. Section 9.8 lets the Board set the detail in the CRIS AI Constitution, overseen by the committee in Article VI. Read the proposed AI Constitution.

How it connects

Awaiting counsel. Article IX is part of a draft under review. It is not legal advice, and it has no effect until the bylaws are validly adopted. Questions about it go to Governance and Compliance.

Next steps

Owner: Governance and ComplianceNext review: 2027-03-01