AI Labs Race to Contain Models’ Bioweapon Risks
Generative AI is moving deeper into the life sciences, helping researchers interpret pathogen studies, troubleshoot laboratory procedures and design proteins. Those same capabilities could lower the expertise barrier for developing biological weapons. The 2026 International AI Safety Report said OpenAI’s o3 outperformed 94% of domain experts on virology-protocol troubleshooting, underscoring why model misuse has become a public-health and national-security concern rather than a distant hypothetical.
By September 2026, developers including Anthropic and OpenAI were expanding biological-risk evaluations, refusal training, monitoring classifiers and tiered access controls. Anthropic activated AI Safety Level 3 safeguards when it released Claude Opus 4 in 2025, while industry groups have also backed red-team exercises and stronger screening of synthetic-DNA orders. The measures aim to preserve legitimate scientific uses while blocking assistance that could enable pathogen acquisition or weaponisation, but executives and biosecurity specialists warn that a single failed safeguard could have catastrophic consequences.
All Coverage
1 original reportsThe Backstory
The history behind this eventAI Leaders Warn Advanced Models Could Enable Deadly Pathogens
Advanced AI is moving deeper into biology and genetic design, offering faster drug and vaccine development while potentially lowering the expertise needed to create toxins, pathogens or biological weapons. That dual-use risk has become a central concern for frontier-model developers including OpenAI, Anthropic and Google DeepMind. The policy challenge is to preserve legitimate scientific gains while preventing malicious users from exploiting models, genomic data and increasingly accessible gene-synthesis services.
Axios reported on July 24 that a MIT FutureTech and University of Queensland survey asked 272 researchers to rank 24 AI risks. Respondents put a 12% probability on dangerous AI capabilities causing catastrophe by 2030 and another 12% on AI-enabled weapons and mass harm; without mitigation, each estimate exceeded 20%. The study defined catastrophe as more than 1 million deaths or $100 billion in damage. In June, OpenAI’s Sam Altman, Anthropic’s Dario Amodei, Google DeepMind’s Demis Hassabis, Microsoft AI’s Mustafa Suleyman and Meta’s Alexandr Wang signed a letter urging Congress to mandate customer and order screening by synthetic DNA and RNA providers.
Subscribe to Mark Radar Weekly
Every Friday, the week's strongest signals in your inbox. Unsubscribe anytime.
If you search news on Google, you can set Mark Radar as a preferred source—our coverage will show up more often in your results. Set as preferred source on Google →