Start with the claim.
Separate what was announced from what the evidence actually establishes.
Independent research by Brennan McCloud
Independent audits of what systems actually establish.
Source-led research on verification, assurance, operations, and the evidence required when intelligent systems make consequential claims or take consequential actions.
A proof can be timeless. Verification cannot.
An independent assurance audit of OpenAI's ten machine-checked advances: strong proof evidence, one different validation boundary, and a missing provenance layer.
Assurance audit · Revision 2Read the findings
Show the boundary.
Make clear what software verifies, what it assumes, and where judgment returns to people.
Leave the receipts.
Expose sources, measurements, caveats, and the path another person can use to check the work.