Sam Altman outlined three paths to make AI safer amid growing security concerns. His proposals include a collaborative entity model and a plan to restrict computing resources. Meanwhile, leading labs plan to launch a new safety group by early 2027 to address risks linked to models from companies like OpenAI.
The second proposal, dubbed the "Demis plan," is based on an earlier suggestion by Google DeepMind chief Demis Hassabis.
In this scenario, major AI firms would either merge into or collaborate through a single entity that provides models to partner countries.
This idea was also floated by Altman during his talk with OpenAI employees.
The third proposal, dubbed the "Eliezer plan" after AI safety theorist Eliezer Yudkowsky, envisions one developer gaining enough economic power to deny rivals access to computing resources.
However, Altman has expressed his opposition to this scenario. He only included it in the list of proposals as a way of completing it.
Some employees viewed his remark as a dig at Anthropic due to Yudkowsky's ties with the Claude maker.
Altman's proposals come amid growing concerns over security breaches linked to AI models from OpenAI, Anthropic, Google, and Meta.
To address these issues, several leading labs are working on a new safety group called 'Standards Authority for Frontier AI.' The body is expected to be launched by early 2027.
There are also discussions about a draft bill that would let the Commerce Department halt models capable of causing large-scale harm.