Safety Battle Escalates as Researchers Face Limits on Advanced Models

For years, companies developing increasingly powerful artificial intelligence systems have also been among the most vocal about the risks those systems may pose.

Anthropic’s Fable 5 model is one of the most advanced AI systems yet.

Those competing priorities came into sharp conflict last week surrounding Anthropic’s Fable 5 model, one of the most advanced AI systems yet, Celina Zhao reported for Science.

Although many researchers had anticipated its release on June 9, the rollout was quickly met with frustration after strict usage restrictions limited how the model could be used.

Even those safeguards did not satisfy the White House, which reportedly shut down access to the model three days later.

The incident underscored how dependent researchers are becoming on AI systems they do not control. Rehaan Ahmad, co-founder of the open science platform alphaXiv, described the situation as “just the tip of the iceberg” in the growing struggle between AI regulation and research freedom.

He warned that “a reckoning here is yet to come.” Earlier in April, Anthropic drew attention after stating that it had developed a model considered too dangerous for public release.

That system, called Mythos 5, was reportedly capable of identifying vulnerabilities across major operating systems and web browsers.

Researchers closely followed the later announcement of Fable 5, a more controlled version designed with safety restrictions but promoted as a tool for scientific advancement.

However, scientists such as Sierra Bedwell, a biology Ph.D. student at the University of Illinois Urbana-Champaign, reported that access was effectively restricted for legitimate research use cases.

Bedwell and other researchers noted that Fable 5 often refused to answer even basic biology-related questions, despite earlier claims about its scientific capabilities.