Anthropic says it blocked AI misuse tied to bioweapons research

Anthropic on Thursday released its most detailed threat intelligence report to date, revealing that its AI models were used by state-sponsored hackers, criminal groups, and researchers pursuing biological weapons-related work over the past nine months — a disclosure that underscores how rapidly AI is reshaping the landscape of both cybercrime and biosecurity risk.

The report, titled "Detecting and Countering Misuse of AI: September 2026," covers activity the company's Threat Intelligence team disrupted between December 2025 and August 2026 across seven categories of harm, including cyber operations, biological misuse, surveillance, influence operations, and conventional weapons development. Claude Haiku, Sonnet, and Opus models were all exploited, though the company said none of the cases involved its newer Claude Fable or Mythos-class models.techxplore+1

Biological Weapons Research Flagged

Among the most alarming findings, Anthropic said it identified attempts to use Claude for research that could support biological weapons development. One case involved a platform serving virologists conducting gain-of-function research on the chikungunya virus, routing queries through US infrastructure to evade regional blocks and using zero data retention services to hide content. Another involved a researcher accessing Claude from an unsupported region to plan mammalian-adaptation work on highly pathogenic avian influenza. A third case involved someone drafting a grant proposal for live orthopoxvirus research — the family that includes smallpox — at what Anthropic described as a state-associated laboratory.theweek+1

"It has long been a concern that AI models might one day reach the level of capability where they can help to make existing pathogens more dangerous — or create entirely new ones," the company stated. Anthropic said its latest models now include stronger safeguards restricting access to dual-use biological research queries.theweek

Cyber Operations Grow More Autonomous

The report also documented a shift in how threat actors use AI for cyberattacks, describing a move from AI as assistant to AI as orchestrator. In one case, an actor linked to the Russian state-nexus group Midnight Blizzard automated nearly every stage of intrusion campaigns targeting Ukrainian government bodies, European diplomatic missions, and drone technology supply chains. The actor used Claude to build phishing infrastructure, modify malware to evade detection, and exfiltrate hundreds of gigabytes of stolen data.anthropic

A separate cluster of financially motivated attackers tied to the ShinyHunters collective used Claude to accelerate mass data theft, stealing millions of payment card records and passenger data across multiple victims. Anthropic noted that AI has "collapsed the labor and tooling gap" between state-sponsored operations and individual hackers, meaning sophisticated attacks no longer require sophisticated attackers.anthropic

Timing Draws Scrutiny

The report's release came two days after an Anthropic researcher publicly resigned, alleging the company and its peers were not acting responsibly with AI development, according to the BBC. "It is instances like this where we need to explore the ethics of it," said Kishan R, an AI researcher based in Noida, India. "How are companies like Anthropic and xAI determining what constitutes safe use of their AI models? This is a democratic question and not a corporate one".bbc+1