Anthropic Restores Fable 5, Mythos 5 Access

https://mexicobusiness.news/cybersecurity/news/anthropic-restores-fable-5-mythos-5-access

Publish Date: 2026-07-01 15:20:00

Author:

Using an unordered list, summarize the following article with between 4 and 8 key points. Anthropic has restored global access to its Fable 5 AI model and resumed limited access to Mythos 5 for selected organizations after the US government lifted export controls imposed in June. Anthropic restored global access to Fable 5 after US export controls were lifted, while launching stronger AI safeguards. The decision follows negotiations between Anthropic, federal agencies, and industry partners, and introduces new safety measures, a proposed industry framework for AI jailbreaks, and expanded government collaboration on frontier AI security.The company announced that Fable 5 would be available globally beginning July 1 through Claude Platform, Claude.ai, Claude Code, and Claude Cowork. Meanwhile, Mythos 5 access was reinstated for a subset of US organizations participating in Project Glasswing. Anthropic says it is coordinating with the government to expand access to additional domestic and international partners.Export Controls and Safeguard DisputeThe Commerce Department imposed export controls on June 12 after Amazon researchers reported a method for bypassing Fable 5’s cybersecurity safeguards. According to Anthropic, the issue involved prompting the model to identify software vulnerabilities and, in one case, generate code demonstrating how a vulnerability could be exploited.In response, Anthropic says its testing found that multiple less-capable models, including Claude Opus 4.8, GPT-5.5, and Kimi K2.7, could identify the same vulnerabilities. The company also says all tested models could produce the same exploit demonstration referenced in Amazon’s report.Anthropic argues that the reported technique did not expose Mythos-level offensive capabilities and represented a “borderline case” within Fable 5’s expanded safety margin, which was designed to block some potentially benign cybersecurity requests as a precaution.To address the issue, Anthropic trained a new safety classifier in collaboration with government agencies and Amazon. The company says the updated classifier blocks the reported bypass technique in more than 99% of cases and that blocked requests are automatically redirected to Claude Opus 4.8.Researchers from the Department of Commerce’s Center for AI Standards and Innovation (CAISI), tested both the previous and updated safeguards and concluded that they were “extraordinarily strong,” according to Anthropic. The company acknowledged that the stronger classifier increases false positives during routine coding and debugging tasks and says it will continue refining the system.Industry Push for Formal AI Jailbreak StandardsBeyond the immediate dispute, Anthropic says the episode exposed a broader problem: the absence of an industry-wide standard for evaluating the severity of AI jailbreaks. The company announced that it is working with Amazon, Microsoft, Google, and other Project Glasswing partners to develop a consensus framework that would score jailbreaks across four criteria:Capability gain beyond existing tools.Breadth of offensive capabilities unlocked.Ease of weaponization.Discoverability of the technique.Anthropic says the framework is intended to help AI developers prioritize mitigation efforts, enable safer deployment of advanced models, and provide governments with a more consistent basis for evaluating risk.The company also announced a new HackerOne program that will allow security researchers to submit potential cybersecurity jailbreaks discovered in Fable 5 for review.Expanded Government CollaborationThe reinstatement of Fable 5 and Mythos 5 comes amid a broader shift in US policy toward frontier AI oversight. US President Donald Trump recently issued an executive order creating a process for frontier AI companies to provide designated government partners with early access to certain high-capability models.Anthropic says it will participate in that process for models that materially advance capabilities relevant to national security. The company also committed to:Providing expanded pre-release access to designated government evaluators.Sharing information on significant jailbreaks and misuse patterns.Contributing threat-intelligence reports through the federal vulnerability clearinghouse established by the executive order.Scaling joint AI security research with government agencies.Supporting a voluntary industry-wide security and evaluation standard.Anthropic says it has worked with the Office of the National Cyber Director, the Office of Science and Technology Policy, the Department of the Treasury, the Department of Commerce, including CAISI, and other national security agencies during the development of the administration’s AI security framework.Commercial Availability and Next StepsUnder the restored rollout plan, Fable 5 will be available globally starting July 1. For Pro, Max, Team, and selected Enterprise plans, the model will count for up to 50% of weekly usage limits through July 7 before transitioning to a usage-credit model.Anthropic says it is working to restore access through AWS, Google Cloud, and Microsoft Foundry as quickly as possible.Mythos 5 remains restricted to vetted organizations, reflecting its classification as a model with cybersecurity capabilities that Anthropic says exceed those of all but the most skilled human security experts. The company says it expects additional discussions with the government and industry partners as the proposed jailbreak framework evolves and as access to Mythos 5 expands to a broader set of domestic and international organizations participating in Project Glasswing.Anthropic says it ultimately wants AI release oversight to operate through a durable and transparent process that applies consistently across frontier model developers, balancing access for cyber defenders with restrictions intended to prevent misuse by adversarial actors.