AI’s Next Test: Acting, Storing, and Going Public

An AI that can operate a computer, a setting that determines whether developers’ requests are stored, and a major AI company preparing to sell shares to the public. These aren’t three versions of a faster chatbot. They’re three developments that could shape how AI gets used, governed, and financed. Here’s what the announcements actually say.

First, action. Google’s Gemini API release notes say its Computer Use tool became generally available on October 30 across both the Interactions and GenerateContent APIs. Google had introduced public-preview support in June, describing actions for browser, mobile, and desktop environments, along with configurable safety policies and prompt-injection detection. General availability matters to developers considering a tool for production applications. But it isn’t a promise that an agent will handle every website or unexpected instruction correctly. If software can take actions on someone’s behalf, deciding which actions need human approval becomes part of the product design.

Second, a quieter change with immediate implications for anyone building on the API. Google’s October 26 release note says new projects on paid tiers using GenerateContent now store requests by default. Existing projects keep their current setting. Developers who don’t want that default can set ‘store equals false’ on each request or turn off GenerateContent logging in the logs-and-datasets dashboard. That distinction matters: this is a change for specified new projects, not a claim that every Gemini conversation is newly stored. It’s also a useful reminder that privacy decisions can live in configuration, not just in a company’s headline policy. Developers handling sensitive material have a concrete setting to check.

Third, money and accountability. On October 26, Anthropic announced that it had filed an S-1 registration statement for a proposed initial public offering of Class A shares. It said it intended to list on Nasdaq under the symbol ANT. The number of shares and proposed price range had not been determined. Anthropic also said the registration statement was not yet effective at the time of its announcement. So this was a filing, not an announcement that the stock had begun trading. A public listing, if it happens, would put a prominent AI developer under a different kind of investor scrutiny—but the filing alone tells us neither its eventual valuation nor when trading might start.

The thread connecting these updates is responsibility at scale. Computer-using AI raises the stakes of a mistaken action. A storage default asks developers to inspect what happens to requests. And an IPO filing begins a process in which claims about a company meet public-market disclosure. The practical takeaway: watch not only what AI can do, but who can control it, what data it keeps, and how clearly its makers explain the risks.