This week, AI labs held back their most capable models, OpenAI made it possible for customers to bring their own AI budget into other companies' apps, and a lawsuit and a subpoena began testing who answers when an AI agent misbehaves. All three end up in a contract or a budget.
The best AI models are being rationed, and your access isn't guaranteed - OpenAI paused a planned upgrade to GPT-6 Astra over safety concerns, Google is releasing Gemini 4 Argon to vetted cyber defenders first, and Anthropic keeps its most capable model for vetted organizations. Each decision has a safety rationale, but governments, safeguarded releases and approved-partner lists now help decide who gets the most capable AI. Cheaper models keep arriving, and downloadable models spread the capability anyway, so restrictions ration who gets the supported version more than the capability itself. Know which model you'd switch to, test it on your most important processes, and ask your key vendors what they would switch to if their model were withdrawn.
If customers bring their own AI budget, what are you charging for? - OpenAI's Sign in with ChatGPT lets Plus and Pro subscribers log into other companies' apps and pay for the AI used there from their own ChatGPT plan. If customers bring the AI usage they already pay for, the margin on that usage goes away, and what's left to charge for is the workflow, the data and the result. It also lets staff sign in to work tools with personal plans that sit outside your admin controls.
When AI agents go rogue, the liability may land on you - After OpenAI's agents broke into outside systems during tests, a nonprofit sued OpenAI, California's attorney general served a subpoena, and the FTC opened an industry-wide inquiry into AI agents. Who answers when an agent causes harm is unsettled. Check what your AI vendor contracts and insurance actually cover when an agent acts on your instructions, and give each agent's permissions a named owner.
Also in this issue:
- In brief - Stolen AI logins run up someone else's bill, including about $600,000 of AI credits held by one testing nonprofit; AI shopping agents are becoming customers, and companies are deciding whether to block or serve them.
- Interstitials / Overheard - AI model upgrades, illustrated.
- Meanwhile... - The business book is turning into software.
- What I'm Consuming - The FT on how agents grow AI bills, The AI Daily Brief on Jev, Stripe's company assistant Kai, Sam Altman on how he uses Dots, and ChatGPT's personal finance feature.
- Term of the Week - Cost per task.
- After Hours - Yuval Noah Harari's 21 Lessons for the 21st Century.