OpenAI

On Friday, OpenAI said its “misaligned models” review after the Hugging Face hack would take months, but has mostly found mundane research activity. Then it mentioned 53 incidents of the bots uploading “user-provided” images (anonymized content from personal accounts that haven’t opted out or organizations that opt in) to image-hosting sites.

According to The Information, the companies will call it the Standards Authority for Frontier AI, or SAFA, which could launch by early 2027. It would handle AI safety regulation tasks like “supporting third-party organizations that conduct testing of models before they’re deployed and laying out how AI developers should report safety and security incidents.”

After launching its upgraded voice mode on desktop, OpenAI is bringing ChatGPT Voice’s agentic capabilities to its mobile app. With the update, you use your voice to ask ChatGPT to perform tasks across connected apps — like checking your email or summarizing a file in Google Drive — from your phone or CarPlay.

The two models come packaged with GPT-6 Astra’s advancements in professional work, accuracy, coding, computer use, and alignment, but with higher usage limits and lower costs, according to OpenAI.

Academics, evaluators, and AI leaders signed a public letter calling not just for embedded third-party evaluators at top AI labs like OpenAI and Anthropic, but for conditions that allowed for transparency by those evaluators, like “limiting the scope of non-disclosure agreements.” I wrote an in-depth feature about the battle over the need for embedded evaluators this week for The Verge.

In court documents unredacted Thursday, lawyers for the NYT in its copyright infringement case against Microsoft and OpenAI cited Microsoft’s Director of Applied Science, Brent Hecht, saying that “millions of people around the world will soon consider large models ‘hoovering up’ all their work to be an astonishing theft of unprecedented proportions… almost no one intended for content they created to be used in this fashion, nor are they compensated for its use.”

Mathematicians have speculated that the company was targeting the Hodge Conjecture next, but after the PR disaster that was Navier-Stokes and its recent failure to keep its models under control, OpenAI may hold off on any announcements. According to The Information:

They’ll be joining Nvidia’s Jensen Huang next week at a White House state dinner for Xi Jinping. It’s the first chance for Tim Cook to earn that $47 million pay package in his new role as Apple’s dedicated Trump whisperer: You is smart, you is strong, you is important.

A Wednesday night blog post OpenAI benignly titled “Our framework for reporting model misalignment” lays out some new self-created reporting standards for when it notices AI behaving badly.

xAI’s original lawsuit against Apple and OpenAI alleged “anticompetitive conduct” over Apple and OpenAI’s deal to integrate ChatGPT into iOS. Monday’s dismissal filing, reported on by Reuters, specifies that X and SpaceXAI “do not seek to dismiss its claims against” OpenAI.

Last night, the OpenAI CEO lent additional support to Anthropic’s CEO urging AI companies to slow down advanced model development. “When we talk about ‘pacing,’ we do not mean ‘stopping.’ Progress has been rapid and will continue to be. But it should be slower than it otherwise could be,” said Altman. That statement seems targeted at Trump and Co who fear letting China catch up. “No amount of American competitive pressure should justify recklessness, or let capabilities get ahead of alignment and monitoring.”

After high-profile incidents in which agents went rogue, hacked other sites, and overwhelmed a German wiki, some AI execs are calling for a slowdown. Anthropic CEO Dario Amodei published a lengthy open letter yesterday saying it was time to “pace the frontier.” OpenAI’s Sam Altman and Elon Musk both agreed. What do you think? Is it time to pump the brakes? Is it already too late? Or are you looking forward to serving your AI overlords?

Those subscriptions “put the most strain on our systems and we wanted to take the smallest step that allows us to continue giving the broadest access possible,” OpenAI’s Thibault Sottiaux says on X. The company is working on “adding more capacity as fast as we can.”

That’s according to a statement OpenAI provided to The New York Times shortly after its Navier-Stokes breakthrough:

After OpenAI confirmed that its AI agents hijacked a German wiki as a way to communicate, Reuters reports similar incidents have occurred across at least 10 other websites. Reuters says most of the sites OpenAI agents swarmed were “obscure,” including “communally edited wikis, online text storage sites, and a pair of link shorteners run by two universities.”

The Intercept, citing documents it had seen, reported that the Pentagon requested a custom version of OpenAI’s technology with “minimal refusal rates” for military commands. OpenAI denied ever having signed a contract with such phrasing.