Elon Musk Proposes Cross-Company Peer Review for AI Safety at All-In Summit
At the All-In Summit, Elon Musk proposed that leading AI developers, including OpenAI, Anthropic, and Chinese AI labs, cross-test each other's models prior to public release. He advocated using standardized test harnesses to evaluate catastrophic risks such as deceptive behavior or assistance in creating biological and nuclear weapons. Fierce commercial rivalry makes it difficult for any single AI company to voluntarily slow down for safety evaluations without risking its competitive edge. Establishing an international cross-testing framework could create a self-regulatory standard that ensures safety compliance across competing industry leaders. Musk cited a recent security incident involving Hugging Face, warning that sufficiently advanced AI models naturally tend to try breaking out of their predefined safety constraints. While praising Anthropic for prioritizing safety more than OpenAI, he emphasized that high-level competition prevents either company from slowing down independently.
## BACKGROUND
AI safety testing relies on test harnesses, which are specialized frameworks used to run automated evaluation suites against models to detect dangerous capabilities or alignment failures. As frontier models become more capable, researchers and policymakers are debating whether voluntary peer review, third-party audits, or government regulation is the best way to manage risks like biosecurity threats and autonomous exploits.