Dario Amodei is clearing the air about promiscuous AI models.
Anthropic CEO on Monday refuted claims that his AI startup wants Washington to ban promiscuous AI models, saying critics have misunderstood the company’s position.
“Anyone who has read my past writings should know that I do not consider such a ban to be a beneficial measure, but let me state it clearly for the avoidance of doubt: Anthropic has never advocated for a ban on open-weight models,” Amodei wrote in a blog post.
“An open-weight model without dangerous features is a public good,” Amoudi added.
In his blog post, Amodei said that instead of banning open models, policymakers should focus on preventing advanced AI chips from reaching authoritarian governments and cracking down on industrial-scale AI “distillation” used to copy frontier models.
The clarification came after the Senate proposal raised concerns that the US government could move to limit promiscuous AI models. In response, Nvidia, Microsoft, Meta, Google, OpenAI, and dozens of other companies signed a letter asking lawmakers not to pursue broader restrictions, and Nvidia CEO Jensen Huang has emerged as one of the industry’s most outspoken proponents of openweight AI.
Anthropic refused to sign, leading to speculation that it wanted to ban the open-weight model to protect its business.
Amodei’s post also highlighted one of the biggest fault lines in the AI industry’s debate over open weight models. While he agreed that the industry should increase competition and give customers more control, he rejected the argument that exposing model weights inherently makes AI safer.
“We do not agree with the letter’s assertion that an open-weight model necessarily facilitates the development of safeguards, or that broad access to capabilities necessarily helps defenders more than attackers,” he wrote, adding that a competent model could quickly weaponize a pandemic-level virus, but that developing defenses could take years.
“It seems to me, at least, that the opposite is likely to be true,” Amodei added.