Tip Sheets
Fable 5 dispute: Cornell expert says jailbreak resistance an ‘unsolved adversarial problem’
June 15, 2026
Media Contact
Anthropic is working to resolve a dispute with the U.S. government after officials imposed export controls on its newest Mythos 5 and Fable 5 artificial intelligence models. The company shut off access to everyone on Friday in order to comply.
Ayham Boucher
Lecturer of information science and the executive director of Cornell’s AI Innovation Hub
Ayham Boucher, executive director of AI Strategy and Innovation at Cornell University, says the debate is being framed as a simple request for Anthropic to “fix the guardrails” and then remove the Fable ban, but that assumes model jailbreaks and guardrails are a solved technical problem.
Boucher says:
“Describing this as a request to ‘fix the guardrails’ makes the problem sound much simpler than it is. That framing assumes model jailbreaks are a solved technical problem and that Anthropic merely implemented the guardrails incorrectly. Jailbreak resistance is not like patching a normal software bug. It is an unsolved adversarial problem.
“Anthropic appeared to know this. That is why Fable required a 30-day data retention period: the safety model depended not only on pre-release guardrails, but also on monitoring, detection and post-deployment mitigation.
“The larger policy question is whether frontier model access should depend on guardrails that everyone knows can be bypassed under some conditions.”
