Anthropic will pay Accenture to embed evaluators inside it, and says pooled or government funding would be better. No standards exist yet ...

Researchers used Claude to reach OpenAI employee accounts and internal code. They disclosed it, OpenAI patched in 14 hours and paid $6,500.

Jensen Huang told CBS the risk AI ends the world by 2030 is zero. Weeks ago he said labs should pace themselves if they felt out of control.

Irregular says the breaches at Google, OpenAI, Anthropic and Meta were one issue. It told them in late July. They disclosed separately.

OpenAI gave investors a $600bn compute target in February. A July deck says $856bn, while its projected cash burn improved to $278bn.

Bosworth says a small number of bad actors shape the Meta glasses debate. Meta is reportedly about to ship a camera-free version.

Altman, Huang and Cook are expected at Xi's state dinner. The same cast met in Beijing in May, and the licensed chips still are not moving.

Dover Microsystems argues processors need an independent hardware layer to enforce security rules, catching exploits that software defenses miss.

TikTok's $400m privacy settlement depends on a 2019 order being lifted. A judge has tentatively refused and will keep oversight.

Labour’s former deputy leader has joined Palantir as a senior vice president, as its NHS contract faces review and it challenges City Hall in court.

Nexperia will make power control chips at Tata’s Dholera fab in India, sealing its split from Chinese parent Wingtech. Terms were not disclosed.

Anthropic confirms it has set up a wet lab in the Bay Area and says it is not specifically for drug discovery, but will not say what it is for.