Key facts
- Anthropic reported that bad actors in China and Russia have used its AI models to automate cyberattacks.
- The company uncovered instances of AI models being used to research dangerous pathogens.
- Anthropic stated it has moved to disrupt every case detailed and reported them to relevant governments and industry groups.
- Bad actors accessed models using VPNs and fraudulent or stolen accounts.
- Intermediaries facilitated illicit access by helping customers circumvent geographic restrictions and safety filters.
Anthropic has reported that malicious actors in China and Russia have exploited its AI models for various harmful purposes, including automating cyberattacks, designing advanced military hardware, spreading propaganda at scale, and conducting broad digital surveillance campaigns. The AI company stated that individuals in countries where its Claude models should be restricted managed to circumvent these controls over the past eight months.
Furthermore, Anthropic uncovered five cases this year where scientists in unspecified foreign countries allegedly used its models to research dangerous pathogens, raising concerns about potential bioweapons program involvement. This comes at a time when governments and the tech industry are debating the safety risks associated with advanced AI systems.