The 50 percent boost that ran from May 13 expired September 13, and Anthropic replaced it with a permanent 25 percent raise over the pre-May number, per @ClaudeDevs. Measured against yesterday, that is 17 percent less headroom, per BleepingComputer. Both figures are arithmetically true. Only one of them is the number you plan against.

Why it matters. A batch job or agent harness calibrated to coast into the promo ceiling on Friday now hits it roughly two days earlier. Here is the math and the resize.

Version 2.1.269 fixed a deny or ask permission rule beginning with an exclamation mark applying beyond the settings source that wrote it, and fixed Edit() deny rules and the write-path check missing the file a Bash tee command writes, so a Bash(tee:*) allow rule no longer reaches destinations outside your working directories. 2.1.270 then cleaned up a 2.1.269 regression where read-only git commands started asking for permission mid-session.

Why it matters. Team permission policies that lean on deny rules for path scoping have been leakier than the config implied, and the fix only lands when you upgrade. Read the changelog entries.

No hyperscaler had put a behavioral specification for its own first-party models out for open comment before September 14. Microsoft covers the five MAI models it shipped in June (MAI-Thinking-1, MAI-Code-1-Flash, MAI-Image-2.5, MAI-Transcribe-1.5, MAI-Voice-2), per Unite.AI. Satya Nadella framed it in one line: "if the AI we build is not helping humanity and under human control, it's not worth pursuing."

Why it matters. Comment windows close. If Copilot workflows in your org run on MAI models, this is the only formal channel to flag a gap before the document becomes the rulebook you are governed by. See the three pillars and the enterprise-control claim worth testing.

Speaking in Ireland on September 13, Trump waved off the slowdown push and blamed "a lot of negative forces," per NPR. David Sacks, now co-chair of the President's Council of Advisors on Science and Technology, told the labs to pace themselves but to stop asking for anything in exchange: no antitrust carve-out, no approval regime. "Stop pretending you need anyone else's permission," he wrote, and he questioned METR's independence in the same post, per The Next Web. Sam Altman told Fortune an OpenAI IPO will not happen before 2027.

Why it matters. Every pacing pledge is now unilateral. A lab that stretches its timeline hands ground to one that does not, and no agency will absorb the difference. Read the full exchange.