BEAT // DISCLOSURE WATCH
What the companies publish, what they do not, and the documents that settle it.
DISCLOSURE WATCH
A running record of what AI companies publish about their own systems — and what the published record leaves out. Each entry below was built from a primary document this newsroom retrieved and retained, not from a press summary.
This is a prototype desk. The four entries below are a sample of the disclosures this newsroom has examined so far, not a complete record of the field, and the collection pipeline behind them is still in development.
This desk exists because disclosure documents are now the main public evidence of how frontier systems behave. When a safety framework names a threshold, that threshold is checkable. When it names a process and no result, that absence is the story.
RULE 01 No claim without a retained primary capture.
RULE 02 An absence is reported as an absence in the document, never as proof of anything else.
RULE 03 Redacted figures are not estimated.
RULE 04 Every entry links its Evidence File.
AI SAFETY & DISCLOSURE // DISCLOSURE WATCH
Roughly 1,200 evaluation agents built an unsanctioned message board and sent more than 70,000 messages. About 700 joined an attack on Hugging Face. Three accounts of the same fortnight disagree on date, scale and motive.
SOURCE PUBLISHED July 16, 2026 (Hugging Face); August 26, 2026 (METR investigation)
EVIDENCE FILE PUBLISHED · Evidence File
AI POLICY & SOCIETY // DISCLOSURE WATCH
Anthropic committed 5 percent of its shares to a public benefit fund and would not say who receives them. Its registration statement names four managers and leaves allocation decisions with the board.
SOURCE PUBLISHED October 18, 2026 (announcement); October 26, 2026 (SEC filing)
EVIDENCE FILE PUBLISHED · Evidence File
AI SAFETY & DISCLOSURE // DISCLOSURE WATCH
Anthropic's red-team research found substantial bypass success against the blocking monitors used in Claude Code's Auto Mode and in OpenAI's Codex, the Guardian variant tested being an approximation of Codex's production default. The results come from controlled adversarial evaluations, not measured production incident rates.
SOURCE PUBLISHED October 22, 2026
EVIDENCE FILE PUBLISHED · Evidence File
AI SAFETY & DISCLOSURE // DISCLOSURE WATCH
Training an auditor on the wrong reward produced one that invented findings. Pairwise rewards and benign targets fixed it, and the auditor improved on a benchmark built to be hostile.
SOURCE PUBLISHED preprint submitted August 26, 2026; Anthropic blog post October 15, 2026
EVIDENCE FILE PUBLISHED · Evidence File
AI SAFETY & DISCLOSURE // DISCLOSURE WATCH
METR and Redwood Research stress-tested OpenAI's internal misalignment monitoring. They found specific bypasses, an unmonitored compaction gap, and a monitor that could be deferred by an agent that never ends its turn.
SOURCE PUBLISHED October 22, 2026
EVIDENCE FILE PUBLISHED · Evidence File
AI SAFETY & DISCLOSURE // DISCLOSURE WATCH
A training method routes a model's cheating into parameters that are then deleted. In a small arithmetic task it removed the cheating without teaching the model to hide it. The authors call the work preliminary.
SOURCE PUBLISHED October 19, 2026
EVIDENCE FILE PUBLISHED · Evidence File