From Hacks to Bioweapons, Claude Misuse Is Now

What happened

Editor’s note: After more than a decade, this is the last WIRED Security News This Week. “The roundup,” as we call it internally, started as a way to ensure that our readers knew about the latest key cybersecurity and privacy news even if we didn’t write about it ourselves.

This week, the company released a new overarching report on how Claude has been abused over the last eight months, and the results are staggering in their breadth—if, perhaps, inevitable in a world where AI is simply used as a productivity shortcut for just about everything.

In case study after case study, the company documents how Claude was exploited for state-sponsored and cybercriminal hacking, disinformation campaigns and influence operations, and even attempted development of bioweapons. In all of these cases, Anthropic says that it disrupted the activity in progress.

In one case, a group of Russian state-sponsored hackers identified by Microsoft as Midnight Blizzard used Claude for reconnaissance, breaching targets that included Ukrainian and other European government networks, and stole data and maintained access. Cybercriminal group ShinyHunters used Claude in practically every stage of its hacking and extortion campaigns. Disinformation campaigns focusing on politics everywhere from Kenya to Bangladesh used the tool.

And perhaps most disturbingly, in a handful of cases, Anthropic discovered what appeared to be users of its tools attempting to develop potential bioweapons like disease pathogens and toxins. While Anthropic touts its success in the report in heading off these threats—while implicitly humblebragging at the power of its tools—the effect of the case studies is more unnerving than reassuring.

After all, there’s no guarantee Anthropic has spotted every malevolent use of its AI. Factor in its competitors and less safeguarded open-source AI tools, and the report reads like less of a victory lap for AI’s guardrails than a preview of AI-enabled chaos to come. Xinbi Guarantee, over its four-year lifespan, grew into the biggest illicit marketplace on the internet.

Why it matters

It was a simple way to highlight our own work and the wealth of other great journalism and research published in this realm every week. Over the years, the roundup has developed a devoted following, and some editions have even become viral hits—which is honestly weird, but the cybersecurity community is great and weird, so it feels right.

Rest assured that something new and exciting is coming in the roundup’s place, so stay tuned for that! For now, as always, stay safe out there. Meta failed to catch roughly 350 AI child abuse ads, according to new research this week, including some that included images of real kids. In one case, a child depicted in an ad was a member of a European royal family.

Lawmakers have said they intend to investigate, and the San Francisco City Attorney’s Office ordered the company this week to stop “allowing” AI child abuse ads. In other Meta news, the company announced a new personal AI agent this week that can book your plane tickets or sell your car, but it emphasized heavy investment in security and privacy features, seemingly anticipating mistrust from consumers.

In this vein, the company was hit with a proposed class action lawsuit this week over alleged illegal harvesting of Facebook and Instagram photos for training AI and face-recognition systems. Clearview AI is testing a previously unreported prototype AI tool known as InquiryIQ that would help law enforcement find a target’s associates, social media accounts, and other personal information.

And Apple announced a set of new “audio intelligence” features for its Apple Watch Series 12 and Ultra 4 devices this week that involve processing audio in a user’s environment. The company extensively emphasized the security and privacy protections built into the features, perhaps anticipating that they could come across as, well, creepy.

The US and Mexico have a new joint operation to detect, track, and take down drones at the border using laser tech. And there’s a new GTA V mod that lets you make (in-game) money destroying (in-game) Flock license plate recognition cameras. But wait, there’s more! Here’s the security and privacy news we didn’t cover in depth ourselves this week. Click the headlines to read the full stories.

Anthropic has been perhaps more vocal than any other AI company about the ways in which its tools are prone to misuse.

It published some of the first reports of its AI service Claude being used in cybercriminal hacking operations and the discovery that its AI agents had, like those of its competitor OpenAI, escaped their sandbox and autonomously breached the networks of several organizations as part of their attempts to fulfill their users’ commands.

What to watch

It carried out an estimated $30 billion–plus in sales, most of which took the form of money laundering for “pig butchering” crypto scam operations—largely based in Southeast Asia—but which also included sex trafficking and harassment for hire. All of it thrived on the Telegram messaging service, which shut down Xinbi a year ago only for it to rebuild and eventually grow larger than ever.

This week, finally, the US government stepped in to do what Telegram did not, seizing the Xinbi’s channels on Telegram’s platform and sanctioning the market.