Researchers furious over Anthropic's hidden AI limits

Anthropic's new Mythos-based models were deliberately made less helpful when they detect users working on AI research, the company's system cards for Mythos 5 and Fable 5 say. The firm limited the models' usefulness for tasks related to developing frontier large language models, citing concerns that advanced AI could accelerate the creation of competing models without equivalent safety protections.

Anthropic said these interventions are intentionally invisible to users. Rather than refusing requests or switching models, Mythos may subtly modify responses through techniques such as altering user prompts. The approach drew swift criticism. "Anthropic's latest model will NOT help you if it thinks your ML research/ML engineering is interesting, and/or will secretly degrade its IQ so that the average engineer won't notice," AI research firm SemiAnalysis wrote.

"mythos will be bad ON PURPOSE on ai 'frontier llm research' tasks, this is very very sad for the research community," Elie Bakouch of Prime Intellect added.

anthropic, mythos 5, fable 5, system cards, model restrictions, ai research, frontier llms, prompt alteration, safety protections, semianalysis