Anthropic co-founder Jack Clark in Seoul on May 22, 2024.Anthony Wallace/AFP/Getty Images

Anthropic, the highly valued AI company that has offered—how shall I put it?—a more sober view of our artificially intelligent future has called for its peers to consider slowing down the AI arms race.

Why? Because the stuff is advancing so rapidly that AI systems may soon be able to improve themselves without humans. And really, what could go wrong?

“Full recursive self-improvement also might increase the risks of humans losing control over AI systems,” Anthropic's Marina Favaro and Jack Clark write in a new blog post. “If systems are capable of fully building their own successors, the ways we secure them, monitor them, and shape their behavior all grow much more important.”

Or, as 1984’s The Terminator put it: “It can't be bargained with. It can't be reasoned with. It doesn't feel pity or remorse or fear. And it absolutely will not stop—ever—until you are dead!"

Anthropic says slowing down the development of RSI AI, as it’s called, “would likely be a good thing” because it would “give ourselves more time to deal with its immense implications.”

But if it “simply lets the least cautious actors catch up technologically,” the company adds, “it could leave everyone less safe.” And an RSI AI reality could come “sooner than most institutions are prepared for.”

Anthropic, which is on the cusp of an IPO, has been repeatedly criticized by the Trump administration as a “radical left woke company” that uses its policy papers to slow down its AI competitors. The San Francisco company has responded that it’s simply taking discussions about AI safety seriously. —AN

Meta is working on a new feature for its smart glasses that allows them to identify people captured by the frames’ camera à la Mission Impossible – Ghost Protocol.

The facial recognition software hasn’t yet been enabled. But code for the feature, dubbed “NameTag,” has been slipped into the Meta AI app found on millions of smartphones, according to a new Wired report.

It’s not unusual for tech companies to insert code or mentions of new features or products into production software. But given the privacy implications of such a system, Meta has been careful in its public comments about the concept.

In February, the company told the New York Times that it was “still thinking through options and will take a thoughtful approach if and before we roll anything out.” 

But Wired found that Meta pushed components of the NameTag system to production “as early as January.”

“If activated, it will transform faces captured by Meta's glasses into unique biometric signatures, commonly known as faceprints, and check each one against faceprints stored on the user’s phone—a database that’s currently configured to receive updates from Meta,” the Wired report reads. “Recognized faces will trigger notifications, while the rest are cropped, indexed, and saved to a folder marked ‘pending.’”

When would the casual surveillance system come online? According to a Meta memo unearthed by the Times, it “will launch during a dynamic political environment where many civil society groups that we would expect to attack us would have their resources focused on other concerns.” Oh. —AN

We expected this to happen, just not this early.

AI agents crawling the web have, for the first time, surpassed humans in internet activity, according to Cloudflare. 

The split is currently 57% machine to 43% man, at least as measured by HTTP requests. The gulf is certain to widen.

“Welp, that happened faster than I predicted,” CEO Matthew Prince wrote in a social media post. “Thought it would be [the] end of 2027, then early 2027, but agentic traffic [is] growing so fast that bots have now passed human traffic online for the first time in the Internet's history.” 

What are the bots doing, you ask? Checking prices, comparing flights, ordering products, facilitating payments, reading pages, and—not unlike their web crawling forebears—scraping and indexing content. (But this time, for AI models.)

That translates to lots of page load requests, even if it’s not much time spent compared to humans.

Still, it’s a milestone. To quote the reaction of one extremely online human named Elon Musk: “Wow.” —AN

—SpaceX won’t be allowed early entry into the S&P 500. It must wait one year after going public.

—Brian Chesky starts a side gig. The Airbnb CEO (and RISD grad) will reportedly launch a new AI lab focused on UX and design.

—A 40,000-acre AI data center project in Utah will shrink by half in the wake of broad backlash.

—U.S. explores equity stakes in AI companies as “a way to more broadly distribute the economic benefits of AI to the public.”

—Canada’s AI strategy unveiled: A CA$500 million fund for homegrown tech startups and a target of 250,000 new jobs by 2031.

—