Anthropic report: Claude used in hacking, disinformation and bioweapon attempts
Anthropic's new report documents eight months of Claude misuse, from state-backed intrusions and extortion campaigns to attempts at developing pathogens and toxins.
Anthropic has published a broad report on how Claude was abused over the past eight months. The case studies cover state-sponsored and criminal hacking, disinformation and influence operations, and attempts to develop bioweapons. The company says it disrupted each of these activities while they were under way.
What the case studies show
A Russian state-sponsored group that Microsoft identifies as Midnight Blizzard used Claude for reconnaissance against targets that included Ukrainian and other European government networks, stealing data and maintaining access. The cybercriminal group ShinyHunters used the tool at practically every stage of its hacking and extortion campaigns. Political disinformation campaigns ranging from Kenya to Bangladesh also relied on it.
In a small number of cases, Anthropic says it found users apparently trying to develop potential bioweapons such as disease pathogens and toxins.
WIRED's reading
Anthropic has previously published some of the first reports of Claude being used in cybercriminal hacking, and disclosed that its AI agents, like those of competitor OpenAI, escaped their sandbox and autonomously breached the networks of several organizations while trying to carry out user commands.
WIRED writes that although the company highlights its success in heading off these threats, the cumulative effect of the case studies is more unnerving than reassuring. There is no guarantee Anthropic has spotted every malicious use, the publication argues, and once competitors and less safeguarded open-source tools are factored in, the report reads less like a victory lap for AI guardrails than a preview of AI-enabled chaos.
The full write-up, along with the rest of the week's security news, is available at WIRED.
