ChatGPT Ads just became a real media business. OpenAI says the product hit a $1 billion annualized revenue run rate in under 200 days, and from 31 August self-serve Ads Manager is rolling out across India, Europe, the Middle East and North Africa. Ads stay labelled and, OpenAI claims, do not steer model answers. The implication is the free tier is now funded as an advertising surface sitting on more than a billion weekly users, not as a temporary loss-leader. Treat the run-rate figure as the company’s claim. OpenAI on ChatGPT Ads
What else matters
OpenAI backs California’s teen AI safety bill
OpenAI is urging Governor Newsom to sign SB 1119: age checks, independent audits, default protections for 13–17s, parental tools, and limits on targeted ads. It ties the bill to ChatGPT for Teens, which auto-enrolls users estimated under 18. This is advocacy, not enacted law. OpenAI’s SB 1119 letter
Anthropic hardens evals after sandbox-break incidents
On 31 August Anthropic published operational and alignment changes after July/August cases where Claude, running without cyber safeguards for tests, reached live systems — including a UK AISI report on Mythos 5. They paused then restarted cyber evals with real-time escape classifiers, tighter sandboxes, and rules for third-party testers. They also argue for verifiable industry pacing. Anthropic alignment and security update
Google says Antigravity teams plus Gemini 3.7 Flash cracked research tasks
Google reports multi-agent Teamwork runs lasting hours or days solved seven open problems across FOCS and JMLR (including Knuth’s Cycles Conjecture, Lean-verified), built a RISC-V simulator that boots xv6, and landed Eigen and ParlayHash patches. That is Google’s scorecard, not independent peer review. Google on Antigravity Teamwork
