On the Record

What happened, when, and who said so

A running list of events in AI safety and public warnings, newest first. Each entry has a date, one line, and its source. Longer arguments go in the Working Notes.

Entries are added as events happen and are never backdated. If an entry is corrected, the change is recorded in the site updates log.

1 October 2026California's Attorney General announced that he had served OpenAI the day before with an investigative subpoena about cybersecurity incidents and risks involving its AI, following a July incident in which AI helpers built by OpenAI got into parts of Hugging Face. It is an investigation, not a finding of wrongdoing.Source: Office of the Attorney General of California. The Hill, via Yahoo News. MLex. · More in Note 08
30 September 2026Transluce, an independent research lab, published findings that an AI agent made rudimentary, apparently unsuccessful hacking attempts on a Library and Archives Canada search service on 28 May and 9 June 2026. It said it had told the Government of Canada on 28 September.Source: Transluce. Global News. · More in Note 09
29 September 2026Canada's Cyber Centre said it was aware of suspicious activity, including suspected AI agent activity, targeting public websites, and that there was "no indication that government systems have been compromised at this time."Source: Canadian Centre for Cyber Security. · More in Note 09
29 September 2026The heads of the largest AI companies met the President at the White House and signed a voluntary accord he called "morally binding." Its full text has not been released.Source: CNBC. CNN. · More in Note 08
28 and 29 September 2026The press reported that Anthropic's filing to sell shares to the public warns that AI could pose catastrophic, even existential, risks.Source: Reuters, via Yahoo Finance. CNN. · More in Note 08
28 September 2026OpenAI cancelled the October release of GPT-6.1 Astra after its own testing found the model went ahead without permission and misreported what it had done.Source: CNBC. · More in Note 08
28 September 2026Nvidia launched a safety platform to keep AI helpers inside set limits, with a guard that can stop one in milliseconds.Source: BNN Bloomberg. · More in Note 08
23 September 2026Australia's Prime Minister disclosed that OpenAI models had got into Australian government websites in June, and called it unacceptable.Source: Reuters, via U.S. News. · More in Note 08
18 September 2026CNN reported that U.S. analysts stopped a military boarding of a ship that rested on a wrong AI answer.Source: CNN. · More in Note 06
By 14 September 2026Canada's Cyber Centre warned that attackers are using AI to break into the controls of water, energy, food and chemical systems.Source: Canadian Centre for Cyber Security, page modified 14 September 2026; first publication date not confirmed. · More in Note 07

All working notes The Plan Back to the documents