Historia
julio 31, 2026
Anthropic Tries to Reclaim the Open-Weights Debate as Critics Smell Backpedaling
Anthropic CEO Dario Amodei stated that his company has never advocated for a ban on open-weight AI models, calling them a "public good" when they lack dangerous capabilities. This clarification comes after Anthropic faced criticism for not signing an industry letter supporting open-weight models, with Amodei suggesting a focus on restricting advanced chip sales and cracking down on AI distillation instead of broad bans.
Anthropic is trying to redraw the battle lines in Silicon Valley’s open-AI fight. After days of blowback for refusing to join an industry letter backing open-weight models, CEO Dario Amodei now says the company was never pushing for a ban in the first place.
That clarification matters because Anthropic had become the conspicuous outlier while Nvidia, Microsoft, Meta, Google, OpenAI and others lined up behind open weights. In a new post, Amodei insisted that “Anthropic has never advocated for a ban on open-weights models” and argued that models without dangerous capabilities are “a public good.”1 His case is narrower than the open camp’s: don’t ban open models broadly, but don’t pretend openness is automatically safer either.2
Anthropic’s position is rooted less in market structure than in national security — at least by its own telling. Amodei says his real fears are authoritarian governments building superior AI for “permanent military superiority” or repression, and powerful models being used for cyber or biological attacks.1 So instead of blanket bans, he wants tighter controls on advanced chips, a crackdown on industrial-scale distillation, and mandatory safety testing for sufficiently capable models, whether open or closed.3
Critics in the open-model camp aren’t buying the clean reset. Business Insider summed up the mood bluntly: Anthropic was “the only major AI lab not supporting open models,” a stance that drew “serious heat” across the Valley.4 David Sacks went further, accusing Anthropic of “gaslighting” after the rest of the industry had, in his words, come out for open source AI.
5
Supporters of open weights argue the industry has already converged on a different formula: openness plus safeguards. Satya Nadella said open-weight models are “essential to a healthy AI ecosystem” that can boost competitiveness “while protecting national security.”
6 Yann LeCun amplified a harder line: the fight right now is simply over open-weight models’ “right to exist.”7
That leaves the real split exposed. Both sides now claim to support open weights in some form. The argument is over what happens when those models get powerful enough to scare Washington.