OpenAI may be close to releasing an AI tool that can take control of your PC and perform actions on your behalf. Tibor Blaho, a software engineer with a reputation for accurately leaking upcoming AI products, claims to have uncovered evidence of OpenAI’s long-rumored Operator tool. Publications including Bloomberg have previously reported on Operator, which is said to be an “agentic” system capable of autonomously handling tasks like writing code and booking travel. According to The Information, OpenAI is targeting January as Operator’s release month. Code uncovered by Blaho this weekend adds credence to that reporting. OpenAI’s ChatGPT client for macOS has gained options, hidden for now, to define shortcuts to “Toggle Operator” and “Force Quit Operator,” per Blaho. And OpenAI has added references to Operator on its website, Blaho said — albeit references that aren’t yet publicly visible.

According to Blaho, OpenAI’s site also contains not-yet-public tables comparing the performance of Operator to other computer-using AI systems. The tables may well be placeholders. But if the numbers are accurate, they suggest that Operator isn’t 100% reliable, depending on the task.

On OSWorld, a benchmark that tries to mimic a real computer environment, “OpenAI Computer Use Agent (CUA)” — possibly the AI model powering Operator — scores 38.1%, ahead of Anthropic’s computer-controlling model but well short of the 72.4% humans score. OpenAI CUA surpasses human performance on WebVoyager, which evaluates an AI’s ability to navigate and interact with websites. But the model falls short of human-level scores on another web-based benchmark, WebArena, according to the leaked benchmarks. Operator also struggles with tasks a human could perform easily, if the leak is to be believed. In a test that tasked Operator with signing up with a cloud provider and launching a virtual machine, Operator was only successful 60% of the time. Tasked with creating a Bitcoin wallet, Operator succeeded only 10% of the time. We’ve reached out to OpenAI for comment and will update this piece if we hear back. OpenAI’s imminent entry into the AI agent space comes as rivals including the aforementioned Anthropic, Google, and others make plays for the nascent segment. AI agents may be risky and speculative, but tech giants are already touting them as the next big thing in AI. According to analytics firm Markets and Markets, the market for AI agents could be worth $47.1 billion by 2030. Agents today are rather primitive. But some experts have raised concerns about their safety, should the technology rapidly improve. One of the leaked charts shows Operator performing well on selected safety evaluations, including tests that try to get the system to perform “illicit activities” and search for “sensitive personal data.” Reportedly, safety testing is among the reasons for Operator’s long development cycle. In a recent X post, OpenAI co-founder Wojciech Zaremba criticized Anthropic for releasing an agent he claims lacks safety mitigations. “I can only imagine the negative reactions if OpenAI made a similar release,” Zaremba wrote. It’s worth noting that OpenAI has been criticized by AI researchers, including ex-staff, for allegedly de-emphasizing safety work in favor of quickly productizing its technology.

Microsoft was once the exclusive provider of data center infrastructure for OpenAI to train and run its AI models. No longer. Coinciding with the announcement of Stargate, OpenAI’s massive new AI infrastructure deal with SoftBank, Oracle, and others, Microsoft says it has signed a new agreement with OpenAI that gives it “right of first refusal” on new OpenAI cloud computing capacity. That means that, going forward, Microsoft gets first choice over whether to host OpenAI’s AI workloads in the cloud — but if Microsoft can’t meet its needs, OpenAI can go to a rival cloud provider.

“OpenAI recently made a new, large Azure commitment that will continue to support all OpenAI products as well as training,” Microsoft said in a blog post. “To further support OpenAI, Microsoft has approved OpenAI’s ability to build additional capacity, primarily for research and training of models.”

OpenAI has blamed a lack of available compute for delaying its products, and compute capacity has reportedly become a source of tension between the AI company and Microsoft, its close collaborator and major investor. In June, Microsoft, under shareholder pressure, permitted OpenAI to ink a deal with Oracle for additional capacity. In the blog post, Microsoft reiterated that “key elements” of its long-standing partnership with OpenAI remain in place through 2030, including its access to OpenAI’s IP, revenue-sharing arrangements, and exclusivity on OpenAI’s APIs. That assumes, of course, that OpenAI doesn’t achieve artificial general intelligence (AGI) under the two companies’ agreed-upon definition before then. When OpenAI develops AI systems that can generate at least $100 billion in profits, Microsoft will lose access to the company’s technology, according to a reported agreement between the firms. OpenAI is said to be considering nullifying the agreement in a possible bid to secure more Microsoft funding. “The OpenAI API is exclusive to Azure, runs on Azure and is also available through the Azure OpenAI Service,” the blog post reads. “This agreement means customers benefit from having access to leading models on Microsoft platforms and direct from OpenAI.” We’ve reached out to OpenAI and Microsoft for more information and will update this post if we hear back.

A Delaware judge has sanctioned Sheryl Sandberg, Meta’s former COO and board member, for allegedly deleting emails related to the Cambridge Analytica privacy scandal. The decision arises from a case Meta shareholders brought against Sandberg and another former Meta board member, Jeff Zients, late last year. The plaintiffs alleged that Sandberg and Zients used personal email accounts to communicate about issues relating to a 2018 shareholder lawsuit that accused Facebook leaders of violating the law — and their fiduciary duties — in failing to protect users’ privacy. Plaintiffs also alleged that Sandberg and Zients deleted emails from their personal inboxes despite being instructed not to do so by a court. In a decision Tuesday, the Delaware judge overseeing the case found the accusations to be convincing.

“The defendants disclosed Sandberg’s personal Gmail account, maintained under a pseudonym, that she used to ‘communicate about matters potentially relevant to the claims and defenses in this action,’” the judge’s decision reads. “Counsel’s failure to give a straight answer in Sandberg’s interrogatory responses or when answering plaintiffs’ questions supports an inference that Sandberg was not using an auto-delete function but rather picking and choosing which emails to delete.”

In sanctioning Sandberg, the judge raised the legal standard for Sandberg’s affirmative defense, the defense based on facts other than those in support of the plaintiff’s claim. Now Sandberg must prove her defense by “clear and convincing” evidence — not merely a “preponderance” of evidence, a burden that’s easier to clear. The judge has also awarded plaintiffs certain expenses. In a statement to TechCrunch via email, a spokesperson for Sandberg said that the plaintiffs’ claims “have no merit.” “All work emails were preserved on Facebook’s servers,” the spokesperson said. At the root of the courtroom battle are allegations that Meta officials violated a 2012 Federal Trade Commission (FTC) order under which the company agreed to stop collecting and sharing Facebook users’ personal data without their consent. Facebook allegedly later sold the data to commercial partners, including political consulting firm Cambridge Analytica; it was also accused of removing disclosures from privacy settings that were required under the FTC’s order. In 2019, Meta agreed to pay the FTC $5 billion to settle charges that the company violated the 2012 order. The company has also paid penalties from regulators in Europe.