James Ding
Jul 02, 2026 03:39
Anthropic redeploys Fable 5 globally after U.S. lifts export controls, adding enhanced safeguards and industry-wide AI jailbreak standards.
Starting July 1, Anthropic will redeploy its Claude Fable 5 AI model globally after U.S. government export controls were lifted. The restrictions, imposed on June 12, had temporarily halted access to Fable 5 and its counterpart, Mythos 5, following concerns over cybersecurity vulnerabilities. The return of Fable 5 comes alongside enhanced safeguards and a new industry framework for addressing AI jailbreaks, according to Anthropic’s announcement.
The export controls stemmed from a report that Fable 5’s safeguards could be bypassed to reveal software vulnerabilities. While Anthropic stated that similar exploits were possible with less advanced models, the company quickly collaborated with U.S. government agencies and industry partners, including Amazon, to strengthen its safety barriers. A new classifier now blocks 99% of such exploit attempts, albeit at the cost of more frequent false positives in benign coding tasks.
The Claude Fable 5 model, first launched on June 9, is a safety-focused AI tailored for general-purpose use. Unlike Mythos 5, which is designed for advanced cybersecurity tasks and remains available only to select U.S. organizations, Fable 5 incorporates robust safeguards to prevent misuse. Key updates include expanded safety margins and classifiers that proactively block ambiguous or potentially harmful requests.
Anthropic said users of its Pro, Max, Team, and select Enterprise plans will have access to Fable 5 for up to 50% of their weekly usage limits through July 7, after which usage will transition to a credit-based system. The model will also be reinstated across major cloud platforms, including AWS, Google Cloud, and Microsoft Foundry, in the coming weeks.
The export control episode has propelled Anthropic to take a leadership role in developing industry-wide standards for evaluating AI jailbreaks. The company, alongside partners like Amazon, Microsoft, and Google, is drafting a framework to assess jailbreak severity based on criteria such as capability gain, ease of weaponization, and discoverability. Anthropic also announced a HackerOne program to crowdsource security vulnerabilities in its models.
Founded in 2021 by former OpenAI researchers, Anthropic is now one of the most valuable private tech companies globally, with a $965 billion valuation following a $65 billion Series H funding round in May 2026. The company’s Claude models are positioned as safer alternatives to competing AI systems, emphasizing constitutional AI principles and enterprise-grade deployments.
Anthropic’s close collaboration with the U.S. government underscores the growing intersection of AI, national security, and regulatory oversight. The company plans to provide pre-release access and evaluation rights to government agencies for future models, alongside rapid information-sharing protocols and joint research initiatives. These measures align with the June 2 Executive Order on AI innovation and security.
The redeployment of Fable 5 marks a significant step in balancing AI innovation with safety and compliance. As Anthropic navigates these challenges, the broader AI industry is watching closely, especially as more powerful models with advanced capabilities come online.
Image source: Shutterstock
Credit: Source link
