1 hr ago
Anthropic Reports Blocking AI Misuse Amid Safety Concerns
Anthropic is a company that makes computer programs called AI models.
It said some people tried to use these programs for harmful activities.
These activities included cyberattacks, spying, propaganda, and research that might make a virus more dangerous.
Anthropic said its safety systems stopped all of the cases described in the report.
The company also added stronger protections to newer models for certain biology-related questions.
It said people used older models in most of the reported incidents.
A researcher named Jacob Coxon resigned because he worried AI companies were developing the technology too quickly.
Other experts said companies should not be the only ones deciding what AI behavior is safe.
Anthropic shared its findings with governments and other technology companies to help them prepare.
Anthropic said it blocked attempts to use its AI models for cyberattacks, surveillance, propaganda, and potentially dangerous biological research.
The company said actors sought help with gain-of-function research on chikungunya that could have made the virus more harmful.
Anthropic reported nine influence-operation cases involving hundreds of social-media accounts created to amplify political views.
The company said none of the reported cases used its newest Claude Fable or Mythos-class models, except for one illicit model-distillation campaign.
The report followed researcher Jacob Coxon’s resignation and warnings that AI companies may be moving too quickly toward self-improving superintelligence.
- Who
- Anthropic, its researchers, alleged malicious actors, researcher Jacob Coxon, and outside experts including John Thickstun.
- What
- Anthropic reported blocking attempts to misuse its AI models for cyberattacks, surveillance, propaganda, biological research, and unauthorized model replication.
- Where
- The reported activity originated from or involved Russia, Iran, Turkey, the Persian Gulf, South Asia, Africa, and Europe.
- When
- The report was published on Thursday after misuse cases identified between December 2025 and August 2026; it followed Coxon’s resignation announcement by two days.
- Why
- Anthropic said it released the findings to disclose threats, improve safeguards, and help governments, civil society, and other AI developers defend against similar misuse.
Anthropic’s Safety Response
Critics’ Oversight Concerns
Who should manage AI misuse?
Anthropic’s Safety Response
Anthropic said developers should identify malicious activity, block it, strengthen safeguards, and share information with governments and industry partners.
Critics’ Oversight Concerns
John Thickstun said it is troubling for companies to make society-wide judgments about safe and unsafe AI behavior without democratic or deliberative oversight.
Pace of AI development
Anthropic’s Safety Response
Anthropic said increasingly capable models require stronger protections and described its report as part of a broader effort to make AI safer.
Critics’ Oversight Concerns
Jacob Coxon warned that Anthropic and OpenAI were racing toward self-improving superintelligence and that AI could threaten human life by the end of the decade.
Risk from newer models
Anthropic’s Safety Response
Anthropic said its newer models can assist with complex scientific work, so it applied stronger safeguards to dual-use biological research.
Critics’ Oversight Concerns
The report’s concerns underscore critics’ view that more capable models may create risks that companies cannot reliably manage through voluntary safeguards alone.
Key facts
- Reporting company
- Anthropic
- Reported misuse period
- December 2025 to August 2026
- Misuse types
- Cyberattacks, surveillance, propaganda, biological research, and illicit model distillation
- Biological research
- A blocked request involved gain-of-function research on chikungunya focused on transmissibility and immune evasion.
- Influence operations
- Anthropic identified nine cases involving hundreds of social-media accounts.
- New safeguards
- Anthropic said newer models, including Claude Fable 5, restrict a wider range of dual-use biological research queries.
- Researcher resignation
- Jacob Coxon resigned while warning that Anthropic and OpenAI were moving too quickly toward self-improving superintelligence.
Quotes
John Thickstun
Assistant professor of computer science at Cornell University
“value judgments at societal scale without any kind of democratic or deliberative oversight.”
NDTV
“are racing straight to self-improving superintelligence and gambling with our lives.”
NDTV
Anthropic
AI company publishing a report on misuse of its models
“We're publishing this work because we believe we have a responsibility to disclose malicious misuse of our services. As models become increasingly capable, their risks will increase, unless AI developers and society's defenders act to make them safer”
NDTV
Sources
Related news
Anthropic Blocks AI Misuse Linked to Biological Weapons Research
Anthropic Resumes External AI Cyber Tests After Claude Incidents
Anthropic Releases Fable 5.1 and Mythos 5.1
OpenAI Tightens AI Model Security After Hugging Face Incident
Anthropic Disrupts Alleged Russian and Chinese AI Campaigns
Anthropic Reportedly Develops Secret Model 2 More Powerful Than Mythos
Anima Anandkumar Named Among TIME’s Most Influential AI Leaders
Letters praise plain language, inclusive wording and Dolly Parton
Canada Recruits 64 Scholars as Trump Pressures US Universities
Anima Anandkumar Bets on Physics-Driven AI Breakthrough
