AI leaves the lab. We follow what happens next.

Archive October 17, 2026 10:12 UTC
Listen to today's briefing 8:45
Signal: Anthropic is drafting its own regulator's veto, borrowing at junk rates for chips, and publishing the hiring its agents replaced - all in one day.
Anthropic Wrote the AI Law. The Bar: 50 Dead.
The draft bill Anthropic has been circulating on Capitol Hill would let Washington halt a model only on a foreseeable, imminent risk of $1bn in damage or 50 casualties - and would move that call to a single federal judge while overriding some state law.
Business Today: only a federal judge could suspend a model, overriding some state rules Storyboard18: evaluators would register with Commerce and flag imminent threats in 72 hours

NEWAnthropic's $60BN Chip Debt Costs 10%
Project North Star asks lenders for $60bn against five years of Google TPU lease payments, with Blackstone offered a 10% coupon on $9bn of junior debt where similar borrowers pay about 7%.
Replit Drops Anthropic: 'ChatGPT Is Better Now'
Amjad Masad told TechCrunch Disrupt that Replit - which he says was one of Anthropic's top five customers - has left on cost, performance and partnership, while people close to the deal insist it is still in place.
Unite.AI: Replit's own docs still list seven Claude models in Max Mode
Workbench Weekly podcast Oct 1 – Oct 8, 2026 · 17:17
NEWSix Agents Per Employee, Fewer Juniors Hired
Anthropic's own report says agent hours passed employee hours in March and ran six to one by August, with relative hiring in junior technical roles slowing - the vendor's data, about itself.
Business Insider: Anthropic staff expect AI doing 80% of their jobs by 2030 Anthropic: the companion report measures the public's expectations gap
Ghost Note Human Neurons in a Dish, Reading Braille