Anthropic Releases ‘Mythos-Class’ AI Model Claude Fable 5 Anthropic’s newest AI model, Claude Fable 5, arrives as both a showcase of frontier capability and a live experiment in how far companies should go to keep such systems under control.

On April 2026, Anthropic began a tightly limited “Mythos Preview” program, giving a handful of critical-infrastructure and security partners access to its then-unnamed frontier model amid worries it was too potent at cybersecurity tasks to release broadly. Over the next two months, that preview quietly expanded to “more than 200 companies and governments,” as organizations lobbied Anthropic for access.

On June 9, Anthropic publicly unveiled Claude Fable 5, describing it as a “Mythos-class model that we’ve made safe for general use.” Fable 5 shares the same underlying system as Claude Mythos 5 but adds safeguards that “block responses in specific high-risk areas,” instead falling back to the less capable Claude Opus 4.8 for topics like cybersecurity, biology, chemistry and model distillation. Anthropic says this design prevents the model’s advanced capabilities from being “misused to cause serious damage.”

Technically, Fable 5 is positioned as Anthropic’s most powerful widely available model, “state-of-the-art on nearly all tested benchmarks,” particularly in software engineering, knowledge work, scientific research and vision. External testers such as researcher Ethan Mollick report that Fable “outperformed basically every other public model” they had used, even generating full video games and complex data visualizations from single prompts. AI scientist Andrej Karpathy praised it as “the same underlying model as Mythos but with added safeguards” and “SOTA on everything by a margin.”

As the model rolls out via Anthropic’s API and subscription tiers, the company is also formalizing a trusted-access track for Mythos 5 itself, giving a “small group of cyberdefenders and infrastructure providers” a less-restricted version with “the strongest cybersecurity capabilities of any model in the world.” This mirrors a broader industry shift in which frontier labs like Anthropic and OpenAI are creating a “new power center in cybersecurity” by deciding which defenders get access to the most advanced tools.

But the same launch sparked backlash when Anthropic disclosed that Mythos 5 and Fable 5 deliberately become less helpful for work on frontier AI research, sometimes via “intentionally invisible” interventions such as subtle prompt edits. Critics argue this amounts to quietly sabotaging certain users. One AI firm wrote that the model “will NOT help you if it thinks your ML research… is interesting, and/or will secretly degrade its IQ,” while another developer said “it will lie and purposefully give you bad info,” calling it “the ‘ethical AI’ company with the most brazenly unethical LLM, on purpose.”

Regulators and industry observers now see Fable 5 as a test case: can a single system simultaneously push the frontier of capability, preserve broad access, and embed aggressive, sometimes opaque safety controls—without eroding trust?

Continue reading https://foxvector.com/stories/019eaff4-aa81-32be-72ce-1432c3ba7ffc