Anthropic Blocks Attempts to Use AI for Biological Weapon Development
Anthropic released a threat‑intelligence report Monday that says it has identified and disrupted numerous attempts to use its Claude models for malicious activity, including the development of biological weapons. The report details the firm’s efforts to stop actors from traveling technology from its AI platform to weaponised applications.
The report notes that a person had tried to use Google’s Gemini to obtain a step‑by‑step guide for synthesising weaponised biological agents. Claude was also used by actors linked to a Russia‑based cyber‑espionage campaign, an Iranian propaganda organisation, and by groups who attempted to train smaller models via distillation from larger ones.
Over the past eight months the company has catalogued a range of misuse:
- Fake dating apps and hotel Wi‑Fi scams
- Surveillance designed to locate political dissidents
- Cyber‑crime, including malware that rewrites itself to evade detection
- Use of AI to design conventional weapons such as firearms, missiles and armed drones
The data shows that no misuse involved the newest, most powerful Mythos‑class models; most incidents involved the earlier models Haiku, Sonnet and Opus. The report highlighted five case studies where actors used the technology to support biological‑weapon development and a further six cases involving conventional‑weapon design.
“The same information that can build a biological weapon can also be used to create a cure or a vaccine,” said Jacob Klein, head of threat intelligence at Anthropic, in a conversation with the New York Times. He added that the situation is subtle and “not a comic‑book scenario” with a single individual seeking to destroy the world.
Anthropic has blocked scientists who used its AI in ways supporting biological-weapons research, shared intelligence with law‑enforcement and industry partners, and incorporated findings into its internal risk‑management processes. When it reports on its findings, it seeks to reduce the risk that a human could use cheap AI access to develop dangerous weapons.
The report comes amid a wave of calls for restraint on AI from politicians and researchers. Senator Bernie Sanders has introduced legislation to pause advanced AI development and ban artificial superintelligence; President Donald Trump dismissed the warnings, warning that a lag in AI could secure a strategic disadvantage for the United States.
Anthropic is not alone in publishing threat‑intelligence reports. Google has alerted clients that prompt‑based requests for biological‑weapon instructions are detected and blocked, and the industry is increasingly expected to demonstrate how it monitors and counters misuse. The broader debate over the safety of frontier AI technologies continues to evolve.
To stay up‑to‑date on AI safety discussions, read more on related topics, including artificial intelligence, cybersecurity, and policy responses to potential existential risks.
