AI groups race to limit bioweapon risks
Anthropic, OpenAI and Google DeepMind are expanding biological-risk testing, trusted-access programmes and monitoring as models become more capable in dual-use science. Unlike cyber testing, end-to-end bio evaluations require slow, costly wet-lab work and can themselves create safety risks, leaving the industry without a clean way to measure the threat.
Why it matters
Biosecurity is becoming a product-access and regulatory constraint for frontier AI, not a side issue. It will shape model segmentation, enterprise controls and the diligence required for AI-biology start-ups, especially where commercial upside and misuse potential depend on the same capabilities.
Read the full article: Financial Times
