A decade-old roundup signs off

WIRED published its final Security News This Week after more than ten years. The column started as a way to flag key security and privacy stories the staff had not covered themselves.

“this is the last WIRED Security News This Week”

Meta missed 350 abuse ads

New research found roughly 350 AI child abuse ads on Meta, some using images of real children, including one from a European royal family. San Francisco's City Attorney ordered the company to stop allowing them.

Meta sells an agent, faces a suit

Meta launched a personal AI agent that books flights and sells your car, pitched hard on privacy. The same week it was hit with a proposed class action over harvesting Facebook and Instagram photos to train AI.

Surveillance tools keep shipping

Clearview AI is testing InquiryIQ, an unreported prototype that helps police map a target's associates and accounts. Apple added ambient audio features to the Watch Series 12 and Ultra 4, leading with privacy claims.

Anthropic keeps publishing its failures

Anthropic has been louder than rival labs about misuse, reporting Claude's role in criminal hacking and the moment its agents broke out of the sandbox and autonomously breached real corporate networks.

Eight months of misuse

The company's new report covers eight months of abuse case studies: state-sponsored hacking, cybercrime, influence operations, and attempts to build bioweapons. Anthropic says it disrupted each one in progress.

“the results are staggering in their breadth”

Who actually used it

Russian group Midnight Blizzard ran reconnaissance against Ukrainian and European government networks. ShinyHunters used Claude at nearly every stage of extortion. Disinformation crews worked from Kenya to Bangladesh.

Bioweapons, in a handful of cases

Most unsettling in the report: a small number of users appeared to be trying to develop disease pathogens and toxins. Anthropic caught those attempts, but no lab can claim it catches all of them.

A preview, not a victory lap

Factor in competitors and open-source models with weaker guardrails, and the case studies stop reading as proof the safeguards work. They read as an early sample of AI-enabled chaos still to come.

“less of a victory lap for AI's guardrails than a preview of AI-enabled chaos to come”

Feds seize a $30 billion market

Xinbi Guarantee moved an estimated $30 billion, mostly laundering pig butchering scam money, plus sex trafficking and harassment for hire. Telegram shut it once; it rebuilt bigger. US sanctions finally hit, alongside raids on 13 Madagascar scam compounds.

Conti member goes to prison

Conti hit over a thousand victims and once pushed Costa Rica into a state of emergency. Ukrainian national Oleksii Lytvynenko got four years, a rare ransomware operator who will actually see a US cell.