Sam Altman outlines 3 paths to make AI safer
What's the story
OpenAI CEO Sam Altman has shared three possible ways to make artificial intelligence (AI) safer as it is developed. He shared these ideas in a message to his employees. His first idea is to create an international organization that makes sure AI developers follow safety rules. It would work much like the International Atomic Energy Agency (IAEA), which oversees nuclear safety. Reports say Altman prefers this idea. Since 2023, OpenAI leaders have supported creating such an organization. They have suggested that it could inspect AI systems, require safety checks, and track how much computing power and energy AI labs use.
Lab merger
Bringing major AI labs together
The second idea is called the "Demis plan." It comes from an earlier suggestion by Demis Hassabis, the head of Google DeepMind.
Under this plan, major AI companies would either join together as one organization or work together through one organization. This organization would provide AI models to countries that partner with it.
Altman also brought up this idea while speaking with OpenAI employees.
Power concentration
The 'Eliezer plan'
The third idea is called the "Eliezer plan." It is named after Eliezer Yudkowsky, a researcher who studies AI safety. Under this plan, one AI developer would become so powerful financially that it could stop competing companies from getting the computing resources they need.
However, Altman has said he does not support this idea. He included it only to make the list of possible approaches complete.
Some OpenAI employees thought Altman's comment was a criticism of Anthropic, the company behind the Claude AI assistant. They thought this because Yudkowsky has connections to Anthropic.
Risk mitigation
Worries about security problems in AI models
Altman shared these ideas at a time when people are becoming more worried about security breaches involving AI models made by OpenAI, Anthropic, Google, and Meta.
To help deal with these problems, several leading AI labs are working together to create a new safety organization called the "Standards Authority for Frontier AI." The group is expected to start operating by early 2027.
There are also talks about a proposed law that would give the Commerce Department the power to stop AI models that could cause harm on a very large scale.