Odaily reported that Anthropic Chief Executive Officer Dario Amodei has taken a firmer public position on artificial intelligence regulation. Amodei said that if a new artificial intelligence model presents specific risks, governments should have the authority to prevent that model from being deployed. His comments appeared in a lengthy article published on Wednesday, in which he focused on how new models should be tested before release and what powers governments should hold when serious risks are identified.
Call for mandatory third-party testing
In the article, Amodei argued that artificial intelligence models should undergo mandatory third-party testing. The purpose of such testing, according to his statement, would be to evaluate risks that the models may create across several different areas. He wrote that if an artificial intelligence system is found to have “unacceptable risks,” then “the government should have the power to prevent or contain its deployment.” The statement marks one of Amodei’s strongest positions so far in favor of tighter artificial intelligence regulation.
Amodei also used comparisons with other technologies that are already heavily regulated. He wrote: “I think, at least in the current phase of exponential development, the most appropriate analogy is cars, airplanes or medicines, technologies that are critical to the modern economy but can also cause mass deaths if designed or used improperly.” By placing artificial intelligence in the same category as these technologies, he framed model deployment as an area where pre-release evaluation and clear limits on use are central issues.
Mythos model and cybersecurity capabilities
Anthropic had previously warned the public about its artificial intelligence model Mythos. According to the company’s earlier warning, Mythos had the ability to discover and exploit critical software vulnerabilities. Because of that capability, Anthropic decided to make the model available only to a small number of partners, rather than opening it more broadly.
This week, Anthropic released a new version of the model and removed the related cybersecurity attack capabilities. The information was relayed by Jinshi, while the newsflash was published by Odaily. Amodei’s latest comments center on the same set of issues: who should have the power to stop a new model from being released, how independent testing should be used, and what role governments should play when an artificial intelligence system is judged to carry unacceptable risks.