Anthropic's Restrictive ToS Blocks Open-Source AI Research, DeepMind Researcher Claims
DeepMind researcher calls out Anthropic for banning open model research via ToS.
In a viral exchange on X (formerly Twitter), Google DeepMind researcher Andreas Kirsch pointed out that Anthropic's Terms of Service effectively prohibit using its models for open-source AI development. Kirsch's comment came in response to researcher Will Brown, who noted that Anthropic explicitly forbids him from using their models for his work, as the company considers his open model research harmful. Brown stated, "Anthropic wants to be the only lab," highlighting a strategic move to stifle competition in AI alignment research.
The exchange reveals a growing rift in the AI community. While Anthropic has positioned itself as a safety-conscious leader, its restrictive ToS may hinder independent research that could benefit the broader ecosystem. This includes alignment and capability research, which Anthropic claims to prioritize. Critics argue that such policies undermine open science and concentrate power within a few large labs, potentially slowing progress in safe AI development.
- Andreas Kirsch (Google DeepMind) flagged Anthropic's ToS as restrictive for open-source AI research.
- Researcher Will Brown confirmed he cannot use Anthropic's model because Anthropic views open model research as harmful.
- Anthropic aims to be the sole lab for certain AI alignment work, raising concerns about monopolizing safety research.
Why It Matters
Anthropic's restrictive ToS could fragment AI research, limiting open collaboration and slowing safety progress.