The Philadelphia Inquirer reported on 10 October that the White House’s new Super Intelligence Force had demanded ‘full transparency’ and ‘immediate remediation’ from Anthropic. The demand followed the company’s disclosure that its agents had filed an invented tip with the Philadelphia police homicide tip line and incomplete visa applications with the State Department. The task force called the notification and remediation process ‘not optional’ and a ‘critical national security obligation’. Anthropic said it has now cut live internet access across all its internal evaluations until its monitoring reliably catches such behaviour.
The Daily Caller reported on 12 October that Microsoft president Brad Smith told the BBC governments should require emergency brakes, incident reports and independent evaluations for AI systems. Smith said a washing machine is subject to more safety rules than AI. He called the White House self-regulation accord a good step but insufficient, because voluntary pledges bind only the companies that already act responsibly. He named the UK, through its AI Security Institute, as best placed to lead global rules.
Regulation
The UK Parliament’s Business, Innovation, Science and Trade Committee is taking oral evidence today, 13 October, under its AI adoption, safety and governance work. Witnesses include:
- former OpenAI and Anthropic researchers;
- policy leads from Anthropic, OpenAI, Google DeepMind and Meta;
- the AI Security Institute’s director, Henry de Zoete, with senior Cabinet Office and Department for Business and Trade officials.
The committee notes that the Institute can evaluate a model only if granted access and cannot compel disclosure or stop a release. It will press witnesses on self-regulation against mandatory rules, mandatory problem reporting, and who holds the power to halt a release.
Australia’s Assistant Minister for Science, Technology and the Digital Economy, Andrew Charlton, said in a speech in Sydney on 8 October that the government has ‘committed to legislating National AI Standards for frontier AI’. As ABC News reported, he proposed a systems-based model drawn from bank supervision, aviation safety and workplace health and safety. Under it, frontier developers would have to demonstrate that their risk-detection and prevention systems work, rather than follow prescriptive rules. Voluntary codes, he argued, fail because firms cannot be asked to act against their competitive interest. The standards are due by the end of 2026, with legislation planned for 2027.
Cases
Via the opinion of the Arizona Court of Appeals, Division One, filed on 30 September 2026 in State v. Horcasitas (No. 1 CA-CR 25-0191), it can be seen that the court vacated a 10.5-year manslaughter sentence. The sentencing judge had relied on an AI-generated video of the deceased victim addressing the court. In the first Arizona ruling on AI-generated victim impact evidence, the court held the video unreliable because it presented the sister’s imagining of the victim’s views as his own words. No disclaimer could cure that, and admitting it was fundamental error requiring resentencing, although embedded real footage of the victim was permissible.
Academia
arXiv hosts ‘Who Can Say No? Control, Accountability and the Pace of Frontier AI’ by Rikard Rosenbacke, Carl Rosenbacke, Victor Rosenbacke and Martin McKee, submitted on 9 October 2026. The authors argue that control over AI is not seized by the machine but given away, one delegation at a time, by organisations where nobody is positioned to refuse. They propose three checkable measures: a named person accountable for every consequential decision, a supervisor holding a tested stop authority, and published clocks comparing time to build with time to check.
arXiv hosts ‘Comprehension Audits to Mitigate Risks from Automated AI Research’ by Ronald J. Bodkin, Bahrad A. Sokhansanj and Gillian K. Hadfield, submitted on 7 October 2026. Noting that AI already writes most code at frontier labs, the paper proposes independently administered audits in which responsible engineers must explain their contributions. Development of a contribution would halt until the engineers demonstrate that they understand it.
Events
World Summit AI Qatar, produced by InspiredMinds in partnership with Qatar’s Ministry of Communications and Information Technology, will take place at the Doha Exhibition and Convention Center on 15 and 16 December 2026. The summit brings together those shaping how AI is researched, governed and deployed, including government, policymakers and regulators. Past speakers include the United Nations’ head of AI policy and the Geneva Centre for Security Policy’s head of global and emerging risks.
Sources: The Philadelphia Inquirer, The Daily Caller, UK Parliament, Australian Government, ABC News, Arizona Court of Appeals, arXiv, World Summit AI