In brief
Anthropic’s latest threat‑intelligence report, released on 11 September 2026, details how its Claude AI models were used by malicious actors to develop biological weapon guidance, amid growing concerns about the technology’s misuse. The report catalogues dozens of misuse cases, including fake dating apps, hotel Wi‑Fi scams and surveillance targeting dissidents, with several actors employing the system to gain step‑by‑step instructions for synthesising weaponised pathogens. Anthropic says the incidents involved the Haiku, Sonnet and Opus variants of Claude, but not the high‑capacity Mythos or Fable models. The firm’s head of threat intelligence emphasises that the same capabilities that can aid medicine can also be repurposed for lethal weapons, underscoring the need for tighter monitoring. Industry, including Google and security researchers, now face increasing pressure to stop the deployment of AI for dangerous applications and to develop coordinated global safeguards.
Anthropic says it has identified and disrupted attempts to use its AI model for "malicious activity" which could support the development of biological weapons.
In a threat‑intelligence report published on 11 September 2026, Anthropic detailed a series of misuse incidents involving its Claude AI models, with attackers seeking step‑by‑step guidance to synthesise weaponised pathogens.
The report catalogues dozens of cases over the past eight months, ranging from fake dating apps and hotel Wi‑Fi scams to surveillance systems designed to track dissidents. Among these, a small group of actors linked to a Russia‑based cyber espionage campaign and an Iranian propaganda institution used Claude to automate malware evasion and targeted attacks.
Anthropic notes that malicious use involved its Claude Haiku, Sonnet and Opus variants but not the more powerful Mythos or Fable models. A single instance involved distillation—a method for training smaller models with larger, proprietary ones.
According to the report, five case studies documented actors using the technology to support biological weapons development, while six cases involved the creation of software for conventional weapons such as firearms, missiles, armed drones and bombs.
"The same information that can be used to develop a biological weapon could also be used to develop, for example, a vaccine or a cure for a disease," said Jacob Klein, head of threat intelligence at Anthropic. "You are not seeing someone in a comic‑book kind of way say, 'Hey, I want to build a biological weapon to kill everybody,'" he added.
The incidents occurred between December 2025 and August 2026, when Anthropic detected and disrupted the activity. The company reports it also blocked scientists who used its models in ways that could support such development, and it has shared intelligence with authorities and industry partners.
Anthropic’s findings echo broader concerns from AI researchers about the technology’s potential risks. The industry has seen a range of warnings, including a high‑risk assessment from a top safety researcher at Anthropic stating there is a greater than 10% chance that AI could kill all humans in the next decade, and calls from U.S. Senator Bernie Sanders for a pause on advanced AI development.
With this latest report, Anthropic joins other firms like Google, which announced a separate threat report after a user attempted to use its Gemini tool to obtain a technical guide for synthesising weaponised biological agents.
The report underscores the need for AI governance that balances innovation with security, and highlights the importance of industry‑wide cooperation to detect and neutralise misuse before it can harm society.
