From Hacks to Bioweapons, Claude Misuse Is Now Everywhere

Security and Privacy Developments This Week Include Reports on AI Tool Usage, Market Enforcement Actions, and Platform Content Policies

Editor’s note: After more than a decade, this edition marks the final installment of WIRED’s Security News This Week roundup. The feature began as a method to inform readers of key cybersecurity and privacy developments reported elsewhere when staff did not produce dedicated coverage. It served to note both internal reporting and external journalism and research in the field. Over time the roundup attracted consistent readership, with certain editions circulating widely within the cybersecurity community. A replacement format is planned.

Meta did not detect approximately 350 advertisements that used artificial intelligence to depict child abuse, according to research released this week. Some of the advertisements incorporated images of actual children. One case involved a child who is a member of a European royal family. Lawmakers stated plans to conduct an investigation. The San Francisco City Attorney’s Office directed the company to cease permitting such advertisements.

Meta also introduced a personal artificial intelligence agent capable of tasks such as booking plane tickets or arranging the sale of a vehicle. The announcement highlighted security and privacy measures. The company faced a proposed class action lawsuit alleging unauthorized use of photographs from Facebook and Instagram accounts to train artificial intelligence and facial recognition systems.

Clearview AI is evaluating a prototype tool called InquiryIQ designed to assist law enforcement in identifying a target’s associates, social media accounts, and additional personal details. Apple introduced audio processing features for the Apple Watch Series 12 and Ultra 4 models that analyze sounds in the user’s surroundings. The company described built-in security and privacy controls for these capabilities.

The United States and Mexico initiated a joint operation employing laser technology to detect, track, and intercept drones along the border. A modification for the video game Grand Theft Auto V permits players to earn in-game currency by destroying in-game depictions of Flock license plate recognition cameras.

Anthropic published a report covering documented instances of Claude usage over the preceding eight months. The report described cases involving state-sponsored actors, cybercriminal groups, influence operations, and attempts to obtain information related to pathogens and toxins. Anthropic stated that it had interrupted the identified activities.

One documented case involved a Russian state-sponsored group tracked by Microsoft under the name Midnight Blizzard. The group used Claude for reconnaissance against Ukrainian and other European government networks, resulting in data access and persistence. The cybercriminal group ShinyHunters incorporated Claude across multiple phases of its operations involving network access and extortion demands. Separate instances involved accounts generating content for political influence efforts in Kenya and Bangladesh. Anthropic reported several accounts that sought details on disease pathogens and toxins. The company attributed these uses to attempts at developing biological agents and stated that access was terminated in each instance.

The report listed specific actions taken by Anthropic in response to each category of activity. It did not claim to have identified every instance of misuse across its platform or those of other providers.

US authorities seized Telegram channels associated with Xinbi Guarantee and imposed sanctions on the marketplace. Xinbi Guarantee operated for four years and facilitated an estimated volume of transactions exceeding 30 billion dollars, primarily related to laundering proceeds from cryptocurrency fraud schemes known as pig butchering operations. These schemes were linked to compounds in Southeast Asia. The platform also handled transactions connected to sex trafficking and harassment services. Telegram had previously removed the service, after which operators reestablished it at larger scale. The Department of Justice announced simultaneous raids on 13 compounds in Madagascar connected to forced labor operations supporting such schemes.

A member of the Conti ransomware group, 44-year-old Oleksii Oleksiyovych Lytvynenko, received a sentence of four years in prison. The group operated until its public disbandment in 2022 and was linked by US authorities to more than one thousand victim organizations and extortion payments totaling millions of dollars. One operation against Costa Rican government systems prompted a declared state of emergency.

Futurism documented accounts on Facebook uploading artificial intelligence-generated videos depicting violence against children, including beatings, burns, confinement, and starvation. Many videos received thousands of reactions from users who treated the content as authentic. The publication located accounts through the platform’s recommendation system after viewing an initial example. Eight accounts were reported through standard channels. Meta removed two accounts, with one removal occurring after an initial rejection. Several review decisions required more than one week. Most videos submitted to the press office were removed, though some remained accessible for a period, including footage showing a child inside a freezer.

Meta’s policies prohibit child sexual abuse material and nonsexual depictions of child abuse in both real and synthetic forms, with stated exceptions for art, cartoons, movies, and games. The policy does not specify the status of artificial intelligence-generated video under those exceptions. In response to the reporting, Meta stated that certain flagged links did not violate its rules.

This concludes the final Security News This Week roundup.