Anthropic has implemented strict measures against users interacting with its AI systems in ways that involve training larger models. A prominent user on X, Jiahao Shao, reported being blocked and having his access to Claude 5.5 downgraded to Claude 5 after attempting to train models on outputs. He claimed he never distilled Claude but used it for job submission and analysis, yet was penalized for "training a competitor" under policies covering RFT and SFT.
Shao criticized Anthropic for hypocrisy, noting it trains on open web data while restricting users from learning from its outputs at scale. He described the anti-distillation stance as a double standard, arguing that protecting a moat shouldn't involve moralizing over scraped knowledge. The incident highlights tensions between open research practices and corporate AI development models, with Shao emphasizing that paying researchers should not be treated as enemies.
His post garnered over 117,100 views, reflecting widespread concern about AI access restrictions and the ethics of data usage in model training.
Source: Hacker News · Summarized by HeadlinesBriefing