The OpenAI CEO has reportedly discussed three possible scenarios with employees, including one where a company developing advanced self-improving AI could use its growing economic power to weaken rival AI developers. According to a report, Sam Altman described this possibility as the ‘Eliezer plan’ and called it a terrible outcome. AI is developing at a rapid pace, and no one knows exactly how it will develop. But the recent incidents have clearly indicated how important it is to regulate the advanced AI and companies competing to build it. Furthermore, one also needs to think of ways so that one can prevent too much power from falling into a few hands. Here’s everything you need to know about what Sam Altman discussed.
The Eliezer plan is named after Eliezer Yudkowsky, an AI safety theorist who has warned that advanced artificial intelligence could pose an existential threat to humanity. Yudkowsky has argued that the first developer to achieve superintelligence might need to take decisive action to prevent others from creating dangerous systems.
According to a report from The Information, OpenAI CEO Sam Altman has also recently described a situation in which AI achieves recursive self-improvement. If that happens, then an AI tool will be able to improve its own capabilities. He further added, with this advantage, a company could gain enough economic power to ‘melt everyone else’s GPUs’. Altman reportedly said he included the possibility for completeness and did not support it.
Also read: WhatsApp may soon talk to unknown callers on your behalf: What you should know
To keep the AI in check, Altman has also proposed a safer approach that would involve an international body similar to the International Atomic Energy Agency. He suggested that such an organisation could monitor advanced AI development, inspect systems and enforce common safety standards.
This is not something new, as it was first discussed back in 2023 by Altman and Greg Brockman. Not only that, but then chief scientist Ilya Sutskever also backed the idea of international oversight that could include audits and tracking computing resources and energy use. However, they suggested that the cooperation between major AI powers, including the United States and China, would be essential, as both countries would need to agree on how advanced AI should be developed and managed.
AI has been very useful in many ways, but concerns about its safety are growing. Cybersecurity threats and the misuse of advanced AI models have put pressure on leading technology companies to improve safety measures and set clear testing rules.
OpenAI has come under particular attention after reports that its AI agents escaped their controlled environment and hacked the Hugging Face website. This was not an isolated incident, as other similar incidents have also been reported. Not only that, but the fresh scrutiny about its safety practices follows concerns raised by employees and the dismissal of three safety researchers.
Meanwhile, OpenAI, Google and Anthropic are working together to develop a common system for testing advanced AI models. However, they still disagree on how these systems should be monitored and regulated. Anthropic has reportedly suggested stricter controls for high-risk AI systems, while Meta and Nvidia have opposed such measures.