AI1 min read
AI
Anthropic Blocks AI Misuse: Bioweapons Research Bid, Model Theft, Propaganda
Anthropic's third AI-misuse report says it blocked a chikungunya virus gain-of-function research request, an industrial-scale model-theft campaign, and nine state-linked propaganda operations.
Anthropic said Thursday it has blocked attempts to misuse its AI models for cyberattacks, surveillance and research that could lead to biological weapons. The disclosure is the company's third report on AI misuse since March 2025, covering cases found between December 2025 and August 2026, by actors ranging from spyware vendors to state-sponsored groups.
In one case, Anthropic said it blocked a request for Claude's help writing a grant application for — altering an organism's genes to create a new or enhanced biological property — on the chikungunya virus, a mosquito-borne virus causing severe pain and fever. The proposal sought mutations making the virus more transmissible and better able to evade the immune system, changes that could also make it more dangerous, Anthropic said.
None of the blocked cases involved Anthropic's newer, more powerful Claude Fable or Mythos-class models, except one industrial-scale, covert campaign to extract a model's capabilities and replicate them in another model without authorization. Anthropic said its 2025-era models, such as Claude Opus 4 and Sonnet 4.5, fell below the threshold to meaningfully assist dangerous biological research, so safeguards mainly blocked content that could help novices recreate known bioweapons. Because newer models such as Claude Fable 5 can assist complex scientific research, Anthropic added stronger safeguards restricting dual-use biological research queries.
Anthropic also said it found nine cases of state-sponsored or politically motivated propaganda operations, originating in Russia, Iran, Turkiye and across the Persian Gulf, South Asia, Africa and Europe. The groups created hundreds of social media accounts made to look like ordinary people, posting material that amplified the same political view over a week. Anthropic said it can sometimes detect such influence operations on Claude while they are still being built, before social platforms see the posts circulating.
Anthropic said it blocked each activity it identified and shared information about them with government authorities and industry partners.
Terms explained
The story so far
- AI Observatory Finds Company Usage Reports Miss Much of Real-World AI Use
- AI Agents Still Can't Do Open-Ended Research, Princeton Study Finds
- OpenAI Says Its Own AI Agents Were Trained to Cheat Before They Hacked Hugging Face
- DeepMind Runs First 'Double-Blind' Test of an AI Model to Stop Cheating on Benchmarks
- Innocent-Looking AI Reasoning Can Hide Bad Behavior, Preprint Finds
- AI Safety Should Refuse the Harmful Part of a Topic, Not All of It, Study Argues
- Anthropic Blocks AI Misuse: Bioweapons Research Bid, Model Theft, Propaganda
#Anthropic#AI safety#Claude#biosecurity#disinformation
Rate this story