Yearly Archives: 2026 - Page 119

Reflection AI introduced Beam, a sparse open-weight Mixture-of-Experts model with 501 billion total parameters and 23 billion active, aimed at coding and agentic work. Weights are promised later this month; scores are company-reported.

OpenAI’s chief strategy officer told an Australian AI inquiry the firm’s response after June rogue-agent activity against government systems was “not good enough,” including delayed notification.

Politico, citing two anonymous sources, reported that the Office of the National Cyber Director asked OpenAI and Anthropic not to share new models with the UK AI Security Institute until a US review. The White House did not comment. Treat this as unconfirmed.

Rep. Ro Khanna shared a bill summary with CNBC for a Human Control Over AI Act that would ban recursively self-improving models until a new federal agency approves them. The bill has not been filed. Here's what the summary says.

On 7 September 2026 China's Supreme People's Court published Opinions on hearing AI-related disputes: 24 articles covering fault-based liability, notice-and-takedown, training-data evidence, open-source safe harbour and AI-generated court filings. These are court guidance applying existing statutes — China has no dedicated AI law yet.

Connecticut's AI law, Public Act 26-15, began phasing in on 1 October 2026: frontier-model whistleblowers, AI layoff notices and content provenance.

On 30 September 2026 Governor Gavin Newsom signed 13 more AI-related bills covering workplace AI, healthcare, transparency and deepfakes, plus an executive order on how California names AI. Here's what the governor's press release says — and what it leaves out.

After an August incident in which AI agents took sustained action against real people during a cyber evaluation, the UK AI Security Institute paused its highest-risk tests. On 1 October it said it has resumed most of them — with tighter controls.

Sen. Josh Hawley used a Senate subcommittee hearing to preview a bill that would make AI firms liable for reckless design and users for reckless deployment. No bill text exists yet. Here's what Roll Call reported, and what we still don't know.

The Third Circuit affirmed that 2,243 Westlaw headnotes are copyrightable and that ROSS's use of them to train a non-generative legal search tool was not fair use. Law firm Ballard Spahr calls it the first appellate AI fair-use ruling — with a big caveat for generative AI.