Anthropic said on Thursday that its Fable 5 language model is once again available to users after the U.S. Commerce Department removed export‑control measures that had halted the service in mid‑June. The company called the original shutdown a “misunderstanding,” referring to a brief clash between its safety safeguards and government concerns about possible jailbreaks and national‑security implications.
Fable 5 debuted on June 9 as a restricted version of Anthropic’s more powerful Mythos 5. The launch featured built‑in guardrails intended to block misuse in cyber‑attacks, while Mythos 5 remained under tighter internal controls because of its broader capabilities. Three days later, officials issued an export‑control directive that barred foreign nationals—including Anthropic employees—from accessing either model. To stay compliant, Anthropic disabled the models for all customers.
The crux of the dispute was whether Fable 5 could be reliably “jailbroken,” a term that describes techniques used to coax an AI system into ignoring its safety limits. Anthropic argued that no evidence showed a universal or easily exploitable jailbreak, and that perfect resistance was technically unattainable. The company warned that demanding absolute jailbreak immunity could stall the rollout of future frontier AI systems.
After a brief review, the Commerce Department lifted the restrictions on June 26, noting that Anthropic had implemented enhanced safeguards. Access is now being restored, and the company is working to bring the model back online for its user base.
The episode underscores a shifting landscape for AI deployments. Historically, new model releases meant incremental improvements in speed, coding assistance, or conversational ability. Today, the most advanced systems are attracting direct government intervention before, during, or after launch. Anthropic’s experience mirrors a similar case involving OpenAI’s GPT‑5.6, which the U.S. government asked to delay a full public rollout and limit early access to vetted partners.
Industry observers see the Fable 5 incident as a preview of how strategic AI technologies may be governed. If regulators continue to treat high‑impact models as sensitive assets, companies may face a new cycle of announcements, temporary shutdowns, negotiations, and reinstatements within weeks. Anthropic maintains that while it supports the government’s authority to block unsafe deployments, any action should follow a transparent, fair, and technically grounded process.
For now, Anthropic is focused on restoring service and reinforcing the protections that led to the initial halt. The company stresses that the model’s safeguards have been strengthened, aiming to prevent the kinds of jailbreak attempts that triggered the export‑control directive. As the AI field pushes the boundaries of capability, the balance between rapid innovation and national‑security oversight will likely become a defining challenge.
Cet article a été rédigé avec l'assistance de l'IA.
News Factory APP - actualités agentiques pour booster votre SEO et AEO.