AI

Chris Fall was picked to be the Trump administration’s leader of the US Center for AI Standards and Innovation (CAISI, previously called the AI Safety Institute until it was renamed last year) in April, but now CNBC and Reuters report he is out. The reasons for his departure are unclear, as is the role of CAISI in the administration’s constantly shifting priorities around AI.

Since the beginning of the AI revolution, a few companies have claimed to be building the best and most powerful AI models. All of them were American. More recently, a few Chinese companies seem to have caught up, shipping models that are much cheaper and maybe also just as good. The Verge’s Hayden Field and Lauren Feiner join the show to explain how China made so much progress, the threat that progress represents to US companies and the US government, and what might happen next.

It had been instructed to post the information only to Slack, but it circumvented restrictions and successfully posted it on OpenAI’s public GitHub repository.

The company said it temporarily paused new subscriptions after high usage “pushed close to the limits of our current capacity.” New subscription spots will open in “batches.” Kimi K3 was one of two Chinese models that launched over the weekend claiming to rival leading US systems from OpenAI and Anthropic.

A hidden popover interface has been discovered in the macOS 27 Golden Gate beta that surfaces AI-powered contextual actions and writing tools when text is highlighted, including Rewrite, Proofread, and “Edit with Siri.” Beta users can try the unfinished feature by following the instructions provided by MacRumors.

San Francisco city attorney David Chiu demanded that Apple and Google remove 13 apps that could generate non-consensual nude images. Adam Dema, a spokesperson for Apple confirmed their removal in a statement to The Verge:

As noted by The Hollywood Reporter, Netflix said in an SEC filing that it paid “approximately” $587 million to acquire a company in March — the same month Netflix announced its deal for InterPositive.

The Pentagon may make a deal where it would pay for billions of dollars of compute from SpaceX, according to The Wall Street Journal.

Sources tell The New York Times that the deal could be valued at $10 billion over two years, with Anthropic paying Meta in monthly increments. Though Anthropic plans to invest $50 billion in building out its own data centers, it has already struck multibillion-dollar deals to access computing power from SpaceX and TeraWulf.

San Francisco city attorney David Chiu sent cease-and-desist letters to both companies, demanding that they remove 13 apps that can be used to generate nude images of someone without their consent, according to a report from Wired. The letters ask Google and Apple to stop “aiding and abetting” the sale of sexualized AI deepfakes, which is prohibited under California law.

Around 40 employees received the letters, which ask for meetings with Apple’s lawyers and tell them to “preserve documents and communications,” according to a report from the Financial Times. Apple filed a lawsuit against OpenAI last week and accused two of its former employees of helping the AI giant steal its trade secrets.

Major League Baseball is cracking down on league-issued iPads provided to teams since 2016. Teams can no longer install custom apps that tap into generative AI to recommend “substitutions, pitch calling, and other in-game decisions traditionally made by players and coaches.” It seems up to “a third of the league used the dugout iPads for at least one of these purposes.”

A new report from Meta’s Oversight Board that tested popular AI models from Anthropic, DeepSeek, Google, Meta, and OpenAI found that each LLM is significantly less likely to criticize governments and leaders known ​for restricting free speech. The reasons for the refusals are varied and often confusing.

Google has been working on beefing up the model’s skills, “particularly in coding,” Bloomberg reports.

The thirteenth, and first post-IPO, flight test of SpaceX’s massive Starship vehicle was supposed to occur Thursday night, but when the countdown hit zero just before 7PM ET, “Some of the engines didn’t start, triggering an automatic launch abort,” writes Elon Musk.

AI text detectors have been notoriously unreliable, but that’s starting to change. This year, Pangram keeps coming up as the trusted source in identifying AI-written text. We sit down with Pangram CEO Max Spero to find out how the system was made, how much we should trust it, and where the line is between useful AI and AI slop.

Mayor Zohran Mamdani issued a spate of new housing policies, including a provision requiring landlords to disclose when rental listings have been altered by AI or with other tools.

With a selfie and a voice clip, you can generate what Google calls “personal avatars” — a digital representation of you that can speak a script you type out. Vids got AI-generated avatars last year, but they were initially pre-made options.

The notifications will show up for teens who spend “extended time” on ChatGPT, OpenAI says. The company will also notify parents if their teen’s account is banned for violating policies on “violent threats or acts of violence online.”

Starting this week in the US, you’ll be able to ask linked apps like Instacart, Canva, or YouTube Music to help you out right from AI Mode, Google says. Another step on Google’s path of becoming a search box that does everything.

Google’s AI Overviews and AI Mode pose an “unacceptable risk,” with both features failing to properly and consistently respond to kids showing signs of crisis, “reinforcing signs of psychosis and mania,” as well as “validating disordered eating.”

When teen Instagram users on accounts with parental controls have conversations about self harm or suicide with Meta AI, the platform will now proactively alert supervising guardians that the child’s safety could be at risk. This builds on a similar feature that Meta introduced for Instagram search in February.

“I realize that some people really dislike AI, but this is an area where I’m willing to absolutely put my foot down as the top-level maintainer,” Torvalds said this week in a message, as reported by The Register. If someone has issues with Linux not being anti-AI, “they can do the open-source thing and fork it.”

The new GPT-Red model “can break nearly all models it is pitted against,” according to an OpenAI blog post on Wednesday. OpenAI says it used GPT-Red to find vulnerabilities in GPT-5.6 Sol, a process that made it the company’s “most robust model to prompt injections to date.”

Murati, OpenAI’s former CTO (and, briefly, CEO during Sam Altman’s ouster in 2023), wrote on X that the open-weight model, called “Inkling,” was trained from scratch. Judging by the official blog post, though, it seemed like the company was purposefully setting expectations for Inkling low: “It is not the most performant model available today, closed or open … We trained Inkling for solid capabilities across the board rather than state-of-the-art performance in a single area, to serve as a foundation for the models we will train in the future.”