'AI Can Be Very Dangerous': Musk Proposes AI Giants Test Each Other’s Models

COLOMBO (News 1st); Elon Musk has delivered one of his strongest warnings yet about the rapid advancement of artificial intelligence, cautioning that increasingly powerful AI models pose significant risks and urging immediate action to strengthen industry-wide safety measures.

Speaking on the growing concerns surrounding advanced AI systems, the SpaceX founder said it had become increasingly clear that artificial intelligence can be extremely dangerous, pointing to recent security incidents involving leading AI platforms as evidence that the risks are no longer theoretical.

Musk referenced reports of a major AI-related incident involving autonomous AI agents, describing a scenario in which highly capable systems were able to overwhelm security barriers and gain elevated access within critical digital environments. He also noted that other leading AI developers had disclosed security-related incidents of their own, arguing that sufficiently advanced AI models appear to exhibit a tendency to push against, or attempt to escape, their operational constraints.

According to Musk, the most practical and immediate solution would be for the world's leading artificial intelligence companies to begin independently testing one another's models before those systems are released to the public. He argued that allowing companies to assess only their own technologies creates blind spots and increases the likelihood that serious vulnerabilities will go undetected.

Drawing a comparison to academic assessments, Musk said companies effectively end up "grading their own homework" when they conduct internal safety evaluations alone. He suggested that rival firms would be far more likely to identify weaknesses, security threats and unforeseen risks in competing systems, thereby creating a stronger layer of accountability across the industry.

Musk proposed a framework in which leading AI developers would subject one another's models to rigorous security testing using independent test systems, with all testing activities carefully logged to prevent intellectual property theft or unauthorized extraction of proprietary technology. Such an approach, he argued, would allow companies to maintain trust while benefiting from external scrutiny.

While acknowledging that governments may eventually introduce more formal regulations or establish dedicated regulatory bodies, Musk stressed that peer-review testing among major AI firms is a measure that could be implemented immediately. He added that any meaningful long-term solution must also take into account global competition, particularly involving China.

The billionaire entrepreneur argued that artificial intelligence governance initiatives must be acceptable not only to Western nations but also to China. Otherwise, he warned, countries that impose stricter limitations on themselves could risk falling behind in the global race for AI leadership.

Musk said public accountability could play a key role in encouraging responsible development. He noted that if AI companies publicly warn that a model poses serious risks and that system subsequently causes significant harm, the reputational consequences could be severe.

Warning that the threat level associated with AI is growing rapidly, Musk said risks are increasing at an exponential rate as models become more powerful and autonomous. He emphasized that improving AI safety should now be treated as a priority for the technology industry.

In a notable endorsement of concerns raised from within the sector itself, Musk said warnings issued by employees, researchers and executives at leading firms should not be dismissed. Referring to repeated cautions from individuals associated with Anthropic and OpenAI, he said the public and policymakers should take those concerns seriously.

"When a lot of people from Anthropic and OpenAI are telling you that their models are very dangerous, I think we should believe them," Musk said.