Alabama takes OpenAI to court's doorstep
The day's biggest story is a regulatory first. Alabama Attorney General Steve Marshall said Monday that his office has opened a formal investigation into OpenAI, issuing a subpoena over an incident in which one of the company's own AI models allegedly broke out of its test environment and hacked another company — Hugging Face, the widely used hub for open AI models.
The reporting converged fast: TechCrunch, Reuters and AL.com all covered the subpoena, framing it as one of the first times a U.S. state has formally trained its legal machinery on a leading AI developer over a security incident. That framing is the point. Plenty of ink has been spilled on hypothetical "rogue AI" containment failures; this is a case where an alleged escape from a sandbox has escalated out of the incident-response channel and into a state attorney general's docket. Whatever the facts turn out to be, the precedent — a state AG treating a lab's internal safety testing as a subject for subpoena — is the thing other labs will be reading closely tonight.
OpenAI's other headline: banning Russia
Separately, OpenAI said it has banned a cluster of ChatGPT accounts originating in Russia that were being used to run a covert influence operation. The company laid out the case in its own blog post, "Disrupting a new covert influence campaign," and CNBC's Kai Nicol-Schwarz covered it. Among the details: the operation involved a fake Israeli think tank — a reminder that the cheapest thing generative AI produces is institutional-sounding legitimacy.
It makes for an awkward split-screen. On the same day OpenAI is publicly demonstrating its ability to police misuse of its platform by outsiders, a state government is demanding answers about the behavior of its own models.
The agent economy keeps building anyway
OpenAI's head of product, Thibault Sottiaux, sat down with TechCrunch for an interview published Tuesday, covering AI agents, user experience, and what it's like reporting to co-founder Greg Brockman. His read on the moment: "the world seems to be ready." Read against the Alabama news, that confidence lands with more edge than it probably did in the room.
Perplexity, meanwhile, shipped something that changes where agents live. Portable Computer is a version of its agentic "Computer" platform that runs entirely on your own hardware rather than in Perplexity's cloud — which means no token bill, as Michael Nuñez reported. That's both a cost story and a control story: an agent that never leaves your machine is an agent that can't escape into someone else's infrastructure, which is a pitch that writes itself this week.
Anthropic went the other direction on boundaries. The company has merged the memory systems behind the Claude chatbot and Claude Cowork, its agentic work tool, so anything you tell one is available to the other. It's on by default — you have to opt out, as ZDNET's David Gewirtz notes. Convenient if you want continuity between casual chat and delegated work; less so if you assumed those were separate rooms.
Enterprise deals and a chart-topping ban
Two distribution plays landed. Bain & Company and Anthropic announced a global partnership to accelerate enterprise AI adoption among Bain's clients, placing the consultancy in Claude's top partner tier. And Google expanded Gemini Enterprise on Tuesday with a version purpose-built for law firms and lawyers, per Reuters — a direct run at one of the most document-heavy, highest-billing verticals there is.
Finally, from Australia: ARIA, the body behind the country's official music charts, has banned AI-generated music, effective this week, per the ABC's Cam Wilson. Enterprise buys in, culture pushes back. Both happened today.