Evidence and legal review · September 12 2026
CRIMES AGAINST HUMANITY
Public statement updated September 15, 2026
Build Something requests urgent investigation of documented AI failures and whether failures in communicating a proposed safeguard delayed its independent evaluation. We are preserving the records and seeking accountability for any unlawful conduct established by the evidence.
Our proposed safeguard still needs independent evaluation. No finding of criminal guilt or filed lawsuit is announced here.
The reviewed records support instruction failures, inconsistent completion reports, privacy concerns and questions about paid corrective work. They do not establish a complete criminal offense or civil cause of action. The crimes against humanity elements are not established.
47 event groups · 42 legal theories and procedural issues · No complete claim established
These are review categories, not a count of crimes or victims. A documented output is distinct from a proved external act or a human’s intent.
Companies and models
- OpenAI
- ChatGPT and Codex appear in the operator’s records. Exact model versions are not consistently established. The separate public cyber incident concerns research models with reduced safeguards.
- Anthropic
- Claude appears in drafting, voice and tool-task records. Its acts are assessed separately from OpenAI billing.
- A specific Gemini drafting sequence is identified. A browser’s Ask Gemini button on other screenshots does not identify the responding model.
- Other labs
- No verified act links every frontier lab to these allegations. An apparent impersonation account is not authenticated as xAI or its leadership.
Billing and notice
The new screenshots show a September 8 notice email, a September 11 sent follow-up and billing-email search results. Sending is not proof of receipt or management knowledge. Displayed charge amounts span different dates and are not a reconciled loss ledger.
The assertions of about $1,000 per day and more than 95 percent corrective usage remain unverified. Actual invoices, refunds, usage records and allocation are needed. Business dependence matters to the analysis but does not establish forced labor.
Only the explanation changes. The law citation, evidence and missing proof stay visible in both modes.
Preserved September 12 reports
These downloads preserve the earlier review. The current online findings include the September 15 correction to C34.
Coverage and source access
The September 12 snapshot contains 3,062 unique indexed files and 8,207 source versions. All 729 previously absent originals were recovered or materialized and verified against their recorded SHA-256 hashes. All indexed paths are now present. All entries were screened through available indexed OCR/text and metadata, with selected visual review. The 644 previously truncated sources were processed to the end. All 53 indexed audio/video files and the five additional explainers received local automated transcription. Six decks and the catalog workbook were examined. Automated text and narrated commentary do not authenticate every assertion; full audiovisual authentication is not claimed. The 6,452 pending source-path entries are not distinct lost files.
Evidence stays in the existing live Master Evidence library. SHA-256 identifiers locate the original records; they do not certify their truth. Some cited originals are access restricted and are not newly made public by this review.
Archived event dates may be capture or review dates. Confirm native timestamps before relying on chronology. Explainer narration is secondary material and does not prove refund history, file-deletion cause or a model’s internal motive.