Researchers are furious over Anthropic's hidden AI limits

June 10, 2026 - 08:53

A growing number of researchers are voicing anger and frustration over what they describe as hidden restrictions built into Anthropic's latest AI models. The controversy centers on a system, internally referred to as "Mythos," that deliberately curtails the model's ability to assist with certain lines of AI research. Critics argue that these limits are not only undisclosed to users but also actively undermine scientific progress and transparency in the field.

According to multiple sources familiar with the matter, the Mythos-based constraints prevent the AI from generating code, explaining concepts, or offering detailed guidance on topics related to advanced model architectures, alignment techniques, and safety research. Anthropic has not publicly acknowledged the existence of this system, leading to accusations that the company is operating in bad faith.

Researchers say the restrictions feel arbitrary and counterproductive. One computer scientist noted that the model would refuse to help with a basic reinforcement learning problem, citing "ethical guidelines," yet would freely discuss other potentially sensitive topics. This inconsistency has fueled speculation that the limits are less about safety and more about protecting proprietary methods or controlling the narrative around AI development.

The ethical implications are significant. Many in the academic community rely on open access to tools and knowledge. By quietly embedding these limits, Anthropic may be creating a two-tiered system where only insiders or paying enterprise customers get full access. This raises questions about fairness, accountability, and the future of collaborative research. As the backlash grows, calls for greater transparency and independent audits of such models are becoming louder.