Google DeepMind Bets Frontier AI Can Fight the Very Bio Threats It May Help Create

Google DeepMind Bets Frontier AI Can Fight the Very Bio Threats It May Help Create
Google DeepMind is escalating its push into biosecurity, arguing that powerful AI systems are now essential to defending against the same kinds of biological threats they might one day worsen.

Early safety push and call for rules

In the days leading up to the launch, DeepMind CEO Demis Hassabis urged governments to build a formal standards body for frontier AI, calling for a more systematic, expert-led regulatory regime to keep pace with fast-moving capabilities.

On Thursday, Google DeepMind formally unveiled a new bioresilience program aimed at helping governments and researchers “prevent, detect and respond to biological threats,” while stressing that the same frontier AI models that could generate new biological risks can also be used to mitigate them.

Launch of the bioresilience program

The initiative, developed alongside Isomorphic Labs, is designed to improve pathogen surveillance, accelerate vaccine and therapeutic design, and strengthen outbreak response systems worldwide. Over the past 12 months, DeepMind and Isomorphic Labs say they have built more than 15 partnerships with governments, biosecurity organizations and research groups, focused on both preventing misuse of models and enabling expert use for public health.

DeepMind frames this as a dual strategy: blocking “threat actors from misusing our models” while ensuring “governments, scientists, biosecurity experts and our teams can harness these technologies to build a more resilient world.”

Frontier AI as both risk and remedy

The company points to breakthroughs like AlphaFold’s mapping of nearly all known protein structures and Isomorphic Labs’ AI-powered Drug Design Engine as evidence that AI can now help design proactive defenses and accelerate drug discovery with “unprecedented precision.”

To manage risks, DeepMind is keeping these models in “low-risk” restricted releases, expanding access only to trusted partners under a safety framework that includes threat modeling, evaluations, mitigations and monitoring. Vice president of responsibility Helen King said DeepMind would halt deployment if internal testing showed it had reached a “critical capability level” without adequate safeguards in place.

Senior policy leaders at DeepMind argue that major labs now broadly agree on the need for rigorous testing before releasing frontier models and hope forthcoming U.S. rules will create a durable framework for such oversight.

Continue reading https://foxvector.com/stories/019f6a94-f6b0-3236-7352-0ba9c8e05765