“How AI makes biological research more dangerous”

Key points

• Anthropic reported five potential instances of actors using its models for biological weapons risk.
• Cases occurred between December and August, per the threat assessment release.
• Two cases involved gain-of-function research on dangerous viruses using Claude.
• Experts warn dual-use research and state actors often evade access controls and signals.
• There is a push for safety standards and possible government review of AI models.