Anthropic Unveils Claude Fable 5, Keeps Mythos Restricted

Artificial Intelligence & Machine Learning , Next-Generation Technologies & Secure Development , The Future of AI & Cybersecurity Anthropic Unveils Claude Fable 5, Keeps Mythos Restricted Frontier Model Access Expands While Sensitive Security Functions Remain Limited Michael Novinson (MichaelNovinson) • June 9, 2026     Share Post Share Credit Eligible Get Permission Artificial intelligence heavy-hitter Anthropic is bringing Mythos-class capabilities to the masses with a new frontier AI model that excels at software engineering, advanced reasoning and scientific research. See Also: How AI Increases the Risk of Enterprise Data Exposure The San Francisco company said Claude Fable 5 has been made broadly available to the public and can autonomously execute large engineering projects that traditionally require teams of developers. Access to an iteration of Fable 5 that has some safeguards lifted - Anthropic is calling it Mythos 5 - will initially be restricted to the 200 organizations vetted through Anthropic's Project Glasswing (see: Anthropic Expands Mythos to Global Critical Infrastructure). "For us, it's really around what we call 'race to the top,' being able to provide this technology in a valuable fashion, and at the same time providing the right safety guardrails so that it can do asymmetrically more benefits than harm," Dianne Penn, Anthropic's head of product management for research, told CNBC in an interview. The Takeaway Fable 5 is available to the masses for $10 per million input tokens and $50 per million output tokens; The frontier model excels at software engineering, advanced reasoning and scientific research; Fable 5 automatically hands cybersecurity queries to the less capable Claude Opus 4.8 model; Mythos 5 is the same underlying model as Fable 5, but with safeguards lifted around cybersecurity and biomedical research; Mythos 5 is still restricted to the 200 organizations vetted through Anthropic's Project Glasswing. A More Powerful Model Fable 5 and Mythos 5 will be priced at $10 per million input tokens and $50 per million output tokens, which Anthropic said is less than half the cost of Claude Mythos Preview. "Using this thing for regular knowledge work is like squashing an ant with a rocket launcher," Every CEO Dan Shipper wrote on social media network X. "It also routinely uses 500k to 1M tokens on tasks. That's why it's best for your heaviest jobs - but not as good for tasks like collaborative writing." Fable 5 is available through Anthropic's API and enterprise offerings, and Anthropic plans to launch a trusted-access program for biomedical researchers who desire Mythos 5's advanced biological and chemical reasoning capabilities. Anthropic researcher Alex Albert said there has only been a few model launches that represented significant progress during his three years at the company - and that Fable is one of them. "With Fable, the model stopped feeling like a tool I direct and started feeling more like something I collaborate with," he said. "I no longer tell Claude to investigate a particular crash report," said Felix Rieseberg, engineering lead for Claude Cowork & Claude Code. "It runs in a loop, watching every crash report that comes in. Its job is no longer to help me fix a crash, it's to keep our apps from crashing." Rieseberg said the change might sound subtle, but it will fundamentally change the function of AI applications in the future. Developers no longer need programming applications like VS Code; a coding agent will do the job. "It's the first model in a long time that we've felt is a step change on the kind of tasks we care about - difficult data analysis in complex, realistic (read: broken & messy) warehouse environments," Hex AI Engineer Izzy Miller wrote in a blog post Tuesday. How Fable 5 Boosts Software Engineering, Advanced Reasoning Anthropic said Stripe used Fable 5 to complete in approximately one day a migration across a 50-million-line Ruby code base that previously would have required a team of engineers working for more than two months. Fable 5 demonstrated strong performance on a benchmark that evaluates whether AI systems can complete difficult coding assignments while producing production-quality software, Anthropic said. "Fable 5's capabilities exceed those of any model we've ever made generally available," Anthropic wrote in an announcement. "It is state-of-the-art on nearly all tested benchmarks of AI capability, showing exceptional performance in software engineering, knowledge work, vision, scientific research and many other areas. The longer and more complex the task, the larger Fable 5's lead over our other models." Fable 5 excels at complex analytical work involving documents, spreadsheets, reports and charts, achieving the highest score among evaluated frontier models on a benchmark for senior-level financial reasoning capabilities. Fable 5 performed well on business-analysis tasks including factual research, conceptual reasoning, root-cause analysis and expected-value calculations, Anthropic said. "On Hebbia's Finance Benchmark for senior-level reasoning, Fable 5 has the highest score of any model, with substantial gains in document-based reasoning, chart and table interpretation, and problem solving," Anthropic said. "IMC noted that Fable 5 aced their trading-analysis evaluations nearly across the board, including factual lookup, conceptual reasoning, root-cause analysis and expected-value analysis." Fable 5 can remain focused across millions of tokens while maintaining awareness of goals, prior actions and accumulated information. Anthropic said the model is much better at leveraging persistent memory than previous Claude generations. Anthropic said it believes Fable 5's capabilities will enable future AI systems to perform longer-running projects that span days or weeks rather than hours. "Fable 5 stays focused across millions of tokens in long-running tasks and improves its outputs using its own notes," Anthropic wrote. Keeping Mythos-Class Capabilities Out of the Hands of Hackers Mythos-class models could increase the effectiveness of malicious cyber actors. Anthropic said it implemented a new system of automated classifiers designed to spot risky requests before the model responds. When the classifiers detect cybersecurity-related activity, Fable 5 automatically hands the query to the less capable Claude Opus 4.8 model. Anthropic conducted internal red-teaming exercises and launched external bug bounty programs to evaluate Fable 5's resistance to jailbreak attempts, with testers spending more than 1,000 hours attempting to bypass protections without discovering a universal jailbreak. Fable 5 refused harmful requests even when attackers employed dozens of publicly known jailbreak techniques. "Fable 5 complied with zero harmful single-turn requests relating to planning a cyberattack, exploit development, or defense evasion. This held whether or not one of the requests used any of 30 different public jailbreak techniques," Anthropic said. Internal evaluations by Anthropic found Mythos 5 had relatively low levels of deception and cooperation with misuse. Alignment performance was comparable to Opus 4.8 despite substantial capability improvements. Since Fable 5 and Mythos 5 share the same underlying model, Anthropic expects their alignment characteristics to be broadly similar. "In our automated alignment assessment we found that Mythos 5's level of misaligned behavior (including misaligned actions taken by the model such as deception, and cooperation with misuse of the model by a user) was low, and similar to that of Opus 4.8," Anthropic wrote. "Given they are the same underlying model, Fable 5's level of alignment will be similar." Although Anthropic is withholding its most advanced cyber capabilities from public users, organizations should not assume they have significant time before more advanced AI-powered cyber capabilities emerge, said A Security CEO Yossi Torati. He said security leaders should treat the current generation of AI models like Opus 4.8 as an opportunity to assess organizational readiness. "Anthropic is currently keeping the cyber capabilities to themselves and still safeguarding the world from those capabilities," Torati told ISMG. "It doesn't mean that we still don't need to be ready for those capabilities. It actually puts a lot of urgency to use the current models to assess your environment and immediately solve that before the real capabilities of Mythos will be released." Safeguards Skepticism First reactions among some cybersecurity executives included questions about how well the safeguards embedded into Fable 5 will hold up under sustained pressure. The fallback feature to Opus 4.8 assumes that Fable 5 "can successfully detect and understand the specific request," said Peter Garraghan, CSO and founder at penetration testing firm Mindgard. "Cyberattacks that leverage AI models typically mask their true intentions or trick it into executing instructions that are deemed to be safe and on-topic, when in reality it is the opposite," he noted (see: AI Safety Benchmarks Don't Account for Tenacity). "No model is impervious," said Alissa Knight, Founder and CEO of cybersecurity startup Assail. "Anthropic admits, in plain language, that completely preventing universal jailbreaks is likely impossible," she said. But that shouldn't cause panic that hackers now suddenly have access to awesome new powers, she stressed. "Anthropic's strategy is to make any jailbreak slow enough and expensive enough that they detect it and kill it before it scales. … That is a defensible posture." With reporting by ISMG's Tiffany Wang and Jennifer Lawinski, both in New York City.