Anthropic’s new fable AI model is met with user backlash over restrictions

Anthropic released its next-generation AI model, Claude Fable 5, on Tuesday, but users and developers are criticizing its aggressive safety restrictions. The model, an update to the previously unreleased Mythos version—deemed too risky for public use due to its potential to exploit software vulnerabilities—now blocks sensitive topics like bioweapons, cybersecurity, and even basic scientific queries such as mitochondria or cancer research. When users attempt to discuss restricted subjects, Fable redirects them to an older, less capable version of Claude or degrades response quality. Researchers, including immunologist Derya Unutmaz from the Jackson Laboratory, report being unable to engage with the model at all, suspecting its filters are overly broad. Unutmaz called the restrictions a ‘nightmare scenario,’ warning that one company dictating permissible questions sets a dangerous precedent. Anthropic defended its approach, stating it initially overblocked biology-related queries to prevent risky research but is now refining safeguards. The company plans to remove restrictions for qualified researchers in biology and life sciences, allowing access to Mythos-class models for biomedical advancements like drug discovery. However, critics argue the degraded performance on AI development tasks undermines transparency and trust in evaluating model capabilities. AI researcher Sayash Kapoor from Princeton University described the backlash as ‘uniform disdain,’ noting this is the first major instance of an AI company implementing guardrails that frustrate users. Others fear the precedent could stifle innovation or force competitors to adopt similar restrictive measures. Despite its limitations, Fable 5 demonstrates improved functionality, with users sharing examples like a basic, functional clone of the video game *Grand Theft Auto* via AI-generated code. The debate highlights tensions between safety and accessibility in AI development, with stakeholders divided over whether companies like Anthropic should unilaterally control model capabilities. While Anthropic works to balance security and usability, the incident has intensified discussions about who should govern AI’s boundaries—developers, governments, or the companies building it.