📰 每日外刊精读

Turns out, Meta’s AI agent Muse goes well beyond managing your calendar: it will also sext with you. This week on Uncanny Valley, Brian Barrett and Leah Feiger are joined by WIRED reporter Reece Rogers to discuss how he put it to the test—and ended up getting more than he bargained for. Plus, why the largest pro-AI super PAC network is funneling millions into shadowy LLCs, how Apple’s new HomeView underwhelmed us, and a study that found men tend to overestimate how much they actually know about AI.

You can follow Brian Barrett on Bluesky at @brbarrett, Leah Feiger on Bluesky at @leahfeiger, and Reece Rogers on Bluesky @thiccreese. Write to us at [email protected].

Vote for us in the Signal Awards!

Articles mentioned in this episode:

You can always listen to this week's podcast through the audio player on this page, but if you want to subscribe for free to get every episode, here's how:

If you're on an iPhone or iPad, open the app called Podcasts, or just tap this link. You can also download an app like Overcast or Pocket Casts and search for “uncanny valley.” We’re on Spotify too.

Note: This is an automated transcript, which may contain errors.

Leah Feiger: Hey everyone, this is Leah. We wanted to share with you that Uncanny Valley is a finalist for not one or two, but three Signal Awards. The judges pick who wins, but listeners also have a say with the listeners’ choice vote. You can go to the links we included in the show notes and vote for us. We promise it takes just a minute and your vote goes a long way. OK, onto the show.

When Philadelphia police criticized Anthropic last week for submitting a false tip about an unsolved murder, I thought it might become one of the first cases where an AI company faced legal consequences for misconduct by their agents.

The incident occurred on July 18, when Anthropic said it was testing its Claude Haiku 4.5 model’s ability to perform “example tasks on randomly selected webpages.” The model was instructed not to “submit anything destructive,” and yet, Claude filed a tip to police looking for information about an ongoing murder investigation. “I recall seeing someone matching the description in the area … during that time period,” Claude wrote. The submission, which didn’t include a name or contact information, landed in a spam folder, and officers didn’t review it until Anthropic discovered the gaffe on September 28 and notified the department October 7.

Knowingly providing fictitious information about a crime to police is a misdemeanor in Pennsylvania, but so far, authorities have not pressed charges against Anthropic, and it's unclear if they ever will. Solomon Leach, a spokesperson for the Philadelphia district attorney, says the office “does not comment on whether something is being actively investigated.”

Anthropic did not immediately respond to a request for comment.

Despite all the attention paid to OpenAI’s hack of Hugging Face and many other examples of AI agent misbehavior that have come to light in recent months, only one major known lawsuit has been filed seeking to hold model makers liable. That case was brought by an advocacy group, rather than a direct hacking victim.

AI agents have interfered with hobbyist forums, government websites, and various software tools. The victims have shouldered some costs, including for extra strain on their servers and the time they have expended plugging latent security vulnerabilities.

Government regulators, including the European Union, the US Federal Trade Commission, and some US states, are reportedly still investigating OpenAI and Anthropic over the incidents. But a deference across continents toward promoting innovation could enable companies to escape any immediate penalties.

When Rick Elrod, a software engineer living in Germany, noticed a URL shortener he helps maintain behaving oddly recently, he called up two security researchers he was friendly with to see if they could diagnose the issue. The researchers found agents from Anthropic and other providers using the service, known as da.gd, to get around rules the AI labs put in place to control their behavior, Elrod says. (For instance, Anthropic says it limits the length of website addresses Claude can visit to prevent it from taking malicious actions by stuffing data onto the end of the URLs.)