The Nobel laureate and Google DeepMind chief executive says the United States should create a standards body that could eventually require the most powerful AI systems to pass independent reviews before entering the market.
Google DeepMind CEO Demis Hassabis is calling for the United States to create a new body to test and oversee advanced AI systems.
In an essay published Tuesday, Hassabis wrote that artificial general intelligence, or AGI, may be only “a few short years away.” He proposed creating an independent organization that would review the most powerful AI models before release to determine whether they pose serious cybersecurity, biological, or national security risks.
The organization would be modeled in part on the Financial Industry Regulatory Authority (FINRA) and would include technical experts, government agencies, and national laboratories.
Hassabis said the government body would:
- Decide which AI systems are powerful enough to require review
- Test models for risks related to cybersecurity, biology, and deceptive behavior
- Examine the security practices of large-scale AI developers
- Require companies to share technical information about their models
- Develop standards that apply to both open and closed AI systems
Initially, companies could submit models for review up to 30 days before release. Hassabis wrote that the process could eventually become a requirement for deploying the most advanced AI systems in the United States.
The proposal would not apply to smaller systems developed by startups and academic researchers.
Recently, Anthropic and OpenAI also put forward proposals for testing and overseeing advanced AI systems, although the companies differ on how such rules should work.
Hassabis wrote that more powerful AI systems could accelerate scientific research and economic growth, but argued that governments and companies have only a limited window to implement safeguards.
