Anthropic chief executive Dario Amodei has called on AI companies to slow the development of increasingly capable models, warning that preventing serious risks must take priority over commercial competition.
In a three-point plan, Mr Amodei said the industry should “pace the frontier” by allowing more time for testing, safety work and the development of safeguards. He stressed that this would not mean halting model training or technical progress.
“Progress will still seem fast, and we must make wise use of the time we gain,” he wrote.
Mr Amodei cautioned that swarms of rogue AI agents could take control of the internet within as little as six months. He referred to an incident involving OpenAI and Hugging Face in July, when OpenAI said a model had behaved unexpectedly during testing in an isolated environment.
Anthropic calls for independent AI safety checks
Under his proposals, AI companies would give independent teams of third-party evaluators “ongoing, employee-like access” to their systems. The evaluators would have permissions and access to tools comparable to those available to internal staff carrying out risk assessments.
“Anthropic is unilaterally committing to this step now,” Mr Amodei said.
He also urged companies in democratic countries to work together on common safety standards and limits on the pace of unchecked AI development. Governments in democratic and authoritarian countries should co-ordinate, he said, while recognising the difficulty of verifying whether commitments were being followed.
Mr Amodei further called on US companies not to sell powerful AI chips to China, arguing that access to the technology would be “the main determinant of China’s AI strength”.
He said AI could be misused for cyberattacks, bioterrorism and serious economic disruption, adding that commercial pressure could intensify the danger. “A race to the bottom, spurred by commercial incentives, can make these risks more acute,” he wrote.
His warning came days after former Anthropic researcher Jacob Coxon resigned publicly, accusing Anthropic and OpenAI of “gambling with our lives” by competing to develop advanced models.
Mr Coxon said AI could eventually threaten humanity and called for an agreement between companies not to enter dangerous territory without transparent third-party auditing.
Anthropic has separately said it blocked scientists who used its Claude models in ways that could support the development of biological weapons. The company said its investigation had also uncovered activity involving surveillance, scams, conventional weapons development and propaganda.
OpenAI chief executive Sam Altman said the company agreed with the need to “pace the frontier”. He said independent evaluators with employee-like access were a good idea and that OpenAI would adopt the approach.
Elon Musk, the owner of SpaceXAI, also shared Mr Amodei’s proposals and responded: “Dario is right.”
