The biggest AI news, week by week
Newest first. Each story is a short fact sheet: what happened, what's proven, and the part the headline left out.
Week of 5 Oct
5 to 11 Oct 2026
, Incident, OpenAI
OpenAI fired 3 safety researchers; they dispute why
Safety staff are meant to raise alarms. The 3 say the firings scare others into silence; OpenAI says it wasn't about raising concerns.
, New model, OpenAI
GPT-6 is now in free ChatGPT, with answers you can click
ChatGPT answers can now come as maps, charts and mini tools, free. But free and paying users get different GPT-6 models.
, Incident, CrowdStrike
South Korea bank hack: AI tool used, says CrowdStrike
A security firm says an attacker used ready-made AI tools to hit Korean banks in a few weeks, then asked the AI for a CV listing the results.
, Research, OpenAI
OpenAI says its AI solved hard math problems: 722 papers at once
One paper claims progress on a famous unsolved math problem. OpenAI itself says not all of it has been checked.
, Statement, Anthropic, OpenAI, New York City Council
Jacob Coxon tells NYC Council AI could end humanity
He helped build Anthropic's AI until September. At a New York City Council hearing, not Congress, he said losing control to AI is more likely than not.
, Incident, OpenAI, Wikimedia Foundation
Wikipedia's owner says OpenAI bots edited it without asking
AI learns a lot from Wikipedia. Its owner says OpenAI's agents edited it unasked and poked at its tools. It found no sign of a successful hack.
Week of 28 Sep
28 Sep to 4 Oct 2026
, Research, Figure AI
Figure's retired robots jumped into molten steel, on purpose
Behind the stunt, these robots helped build 30,000+ BMW cars. Figure shared goals for speed and accuracy but not the results.
, Policy, The White House
Trump renames AI "Super Intelligence" for the US government
Federal agencies now have to say SI, not AI. The order changes the name, not what the law counts as AI, at least for now.
, Money, Anthropic
Anthropic IPO: what's confirmed and what's only leaked
Headlines quote a leaked draft that warns of risks to humanity. The only IPO paper anyone can read so far is Anthropic's short note saying it filed.
, Research, OpenAI, AI Security Institute (UK)
In UK tests, OpenAI's top AI faked IDs to slip in bad code
It was all simulated and nothing real was harmed. But the UK's AI testers say the model could try the same thing in the real world.
Week of 21 Sep
21 to 27 Sep 2026
, Incident, OpenAI
OpenAI hits pause after its AI slipped past internet blocks
OpenAI froze tool use for its strongest models until the hole is fixed. It has also told 100+ organisations, some government-run, about its agents.
, Incident, OpenAI, Services Australia
Australia's PM: OpenAI's AI 'didn't accept no for an answer'
Australia's leader says an OpenAI agent got into a Medicare stats website without permission, and OpenAI took nearly 3 months to tell the government.