From Hacking to Biological Warfare: The Ubiquity of Claude Misuse

In a notable incident, a group of Russian state-sponsored hackers recognized by Microsoft as Midnight Blizzard utilized Claude for reconnaissance, targeting Ukrainian and various European government networks. They successfully breached these targets, stealing data and maintaining ongoing access. The cybercriminal group ShinyHunters leveraged Claude at nearly every phase of its hacking and extortion operations. Disinformation campaigns concerning politics spanned countries from Kenya to Bangladesh, employing the tool. Alarmingly, in a few instances, Anthropic found users attempting to create potential bioweapons, such as disease pathogens and toxins, using its tools.

Although Anthropic highlights its achievements in the report for countering these threats—while subtly boasting about the strengths of its tools—the case studies present a more unsettling picture than a comforting one. Ultimately, there’s no certainty that Anthropic has detected every malicious application of its AI. When you consider its competitors and the risks associated with less secure open-source AI tools, the report reads less like a celebration of AI’s safeguards and more like a glimpse into the looming chaos that AI could unleash.

Xinbi Guarantee, over its four-year existence, evolved into the largest illicit marketplace on the internet, generating an estimated $30 billion or more in sales. The majority of these transactions were linked to money laundering for “pig butchering” crypto fraud schemes, primarily centered in Southeast Asia, along with activities like sex trafficking and harassment-for-hire. This operation thrived on Telegram, which shut down Xinbi a year ago, only for it to reappear and expand further. This week, the US government intervened where Telegram had not, seizing Xinbi’s channels on the platform and imposing sanctions on the marketplace. The Justice Department also announced raids on 13 scam operations in Madagascar, signaling that Western law enforcement is starting to address the widespread issue of forced labor crypto scams, along with evidence of the extensive reach of these operations.

The ransomware group Conti was, before its official dissolution in 2022, one of the world’s most dangerous hacking factions. According to US authorities, it targeted over a thousand victims, extorting millions and at one point causing such significant disruption to government systems in Costa Rica that it prompted a state of emergency. Currently, one member of that group is facing legal consequences: 44-year-old Ukrainian Oleksii Oleksiyovych Lytvynenko was sentenced to four years in prison this week, marking a rare case of a ransomware perpetrator serving time in a US prison.

Facebook is reportedly hosting a vast network of accounts that upload AI-generated videos depicting violence against children, according to Futurism, which spent considerable time documenting this content and continually uncovered more than it could keep track of. These clips portray young children being beaten, burned, confined, and starved. Many draw thousands of reactions from users who seem to believe the footage is real. Futurism indicated that it discovered most of these accounts by accessing one and following Facebook’s recommendation feed, which provided a continuous stream of similar videos—a clear indication that Meta’s own systems can already identify the type of content the company claims to prohibit.

Futurism reported eight of the accounts through the standard user reporting channel. Meta took down two accounts, one of which was initially rejected, and several decisions took over a week. The company deleted the majority of the videos shared with its press office while allowing others to remain live, including a video featuring a child locked in a freezer.

Besides its comprehensive bans on child sexual abuse material, Meta’s policy explicitly prohibits representations of nonsexual child abuse, whether authentic or synthetic, although there are exceptions for art, cartoons, movies, and games. It remains ambiguous whether AI-generated videos fall under these exceptions. In a statement, Meta informed reporters that some flagged links did not violate its guidelines, urging them not to report otherwise.