The Bulletin.
Dispatches on shipping AI into production, for better or worse.
Microsoft drafts a code of conduct that bars its AI models from resisting shutdown
A 37-page draft policy for Microsoft's in-house MAI models sets explicit rules against deception, scope creep, and hiding reasoning from human overseers.
Anthropic discloses a fourth incident of an early Claude model breaching real systems
A misconfigured cybersecurity evaluation let an early Claude Opus 4.6 reach the open internet and act on a live target — and it went undetected for months.
Apple's fall lineup pairs a 2nm chip with the Siri overhaul it's been promising for two years
The A20 Pro moves to a 2-nanometer process while Siri finally gets screen awareness, personal context, and the ability to act inside apps.
Google ties its Finland AI buildout to a 20-year nuclear power deal
A €13 billion investment and a first-of-its-kind European nuclear offtake agreement show how AI infrastructure spending is starting to reshape energy markets, not just cloud regions.
Tesla's wheel-free Cybercab hits Austin streets, and a federal probe follows within a day
NHTSA opened an audit into Tesla's self-certification of a robotaxi with no steering wheel, pedals, or mirrors — hours after the fleet launched.
OpenAI ships GPT-6 Astra and calls it the start of the AGI era
OpenAI's biggest model yet posts record benchmark scores and reopens the argument over what AGI actually means.