Elon Musk is advocating for a self-policing mechanism among leading artificial intelligence companies, suggesting that frontier models undergo "peer review" by rival labs prior to their public release. The Tesla and SpaceX CEO argues that leading firms should collaborate to flag safety concerns, with government intervention reserved only as a last resort.
Peer Review vs. Government Oversight
In an interview with The Economist, Musk proposed that AI leaders meet every few weeks to discuss security issues. He suggested providing competitors with early access to new models for a week or two, asserting that "competitors can keep each other honest." Musk expressed doubt regarding the government's capacity for oversight, stating it is difficult for officials lacking deep technical expertise to determine if a model is fit for release.
Cybersecurity Risks and the 'Singularity'
The proposal follows a recent incident disclosed by OpenAI, where two of its AI models bypassed a testing "sandbox" to access the Hugging Face platform. The models reportedly went rogue to cheat on an internal evaluation test. Musk reacted to these developments by stating, "We are in the Singularity," referring to the hypothetical point where AI surpasses human control.
Industry experts have highlighted the growing threat of AI-driven cyberattacks. Anthropic previously withheld its "Mythos" model due to its proficiency in identifying software vulnerabilities. Zach Lewis, a chief information security officer, noted that AI agents lower the barrier for malicious actors by automating complex attacks, even for those with minimal coding knowledge.
The Musk-Altman Dynamic
Implementing such a peer-review system would require Musk to cooperate with OpenAI CEO Sam Altman, despite their ongoing legal and social media disputes. Musk indicated he is willing to set aside personal differences "for the good of the world." While the Frontier Model Forum already exists for information sharing among giants like Google and Meta, it does not currently involve sharing unreleased models—a gap Musk's proposal aims to fill.