"When Fable's classifiers detect a request related to cybersecurity, biology and chemistry, or distillation, the response is automatically handled by Claude Opus 4.8 instead," Anthropic wrote Tuesday in its launch post for Claude Fable 5. The company is making the Mythos-class model broadly available, but said more than 95% of Fable sessions should stay on the new model while flagged requests fall back to Opus 4.8.

The release makes the control layer part of the product. Safeguards, data retention and trusted-access rules now decide where Mythos-class capability can operate. Pro, Max, Team and seat-based Enterprise users get Fable through June 22; on June 23, Anthropic says usage moves to credits.

Fable 5 and Mythos 5 carry the same price: $10 per million input tokens and $50 per million output tokens. CNBC and TechCrunch described that as twice Opus 4.8; Anthropic said it is less than half the price of Mythos Preview.

Key Takeaways

AI-generated summary, reviewed by an editor. More on our AI guidelines.

What the Opus 4.8 fallback covers

Anthropic's launch post names three classifier categories for Fable 5: cybersecurity, biology and chemistry, and distillation. Users are told when the fallback occurs, the company said, and Opus 4.8 supplies the answer instead of Fable.

Penn framed that choice as a product compromise in her WIRED interview. "Out of all the different approaches, this emerged as the most viable and the best one," she said. Anthropic's own post adds the catch: the safeguards are tuned conservatively and "will sometimes catch harmless requests," even though early data shows fallbacks in less than 5% of sessions on average.

That less-than-5% figure is the operating promise for Fable 5. Most ordinary work should run on the new model. Security teams, biology researchers and anyone testing the boundary will learn how often Anthropic's classifiers move their work to Opus 4.8.

GitHub's data-retention switch

GitHub supplies the clearest enterprise implementation. In a Tuesday changelog, GitHub said Claude Fable 5 is available for Copilot Pro+, Max, Business and Enterprise users, but Business and Enterprise administrators must enable an off-by-default Fable 5 policy.

GitHub tied the switch to retention. The company said Anthropic "retains prompts and outputs for up to 30 days" to operate safety classifiers and deletes them after that period. Retained data is not used to train Anthropic's models, GitHub said. Other Claude models in Copilot, including Opus 4.8, continue to operate under zero data retention.

For an engineering team, the retained material can include proprietary source, architecture notes and bug reports. Anthropic says the 30-day window is needed to detect new jailbreaks and reduce false positives; GitHub's default-off policy for Business and Enterprise shows the tradeoff is material enough that organizations must actively enable it.

Where Mythos 5 stays gated

Mythos 5 is the same underlying model with safeguards lifted in some areas, Anthropic said, and it is initially available to Project Glasswing partners. The company's Mythos page says select biology researchers are next in the access plan, while Tuesday's launch post says the rollout is being done in consultation with the U.S. government.

Know someone who'd find this useful? ✉️ Email it to a friend in one click, or they can subscribe free here.

Project Glasswing started in April with roughly 50 partners and expanded last week by about 150 new organizations across more than 15 countries, per Anthropic and prior Implicator coverage. Anthropic said the restricted model helped partners find more than 10,000 high- or critical-severity vulnerabilities.

Anthropic says trusted defenders need access before attackers get similar capability elsewhere. "The uplift from Mythos-level capabilities is valuable to many adversaries," the company wrote in the Fable post. Justin Beals, founder of Strike Graph, told SiliconANGLE that controlled rollout had "the right instinct" but that "opacity is not a security strategy." Anthropic has not published detailed criteria for Mythos access.

What buyers are paying for

Anthropic is selling Fable 5 as a model for long-running work. Stripe said in Anthropic's launch materials that Fable 5 handled a migration in a 50-million-line Ruby codebase in one day that otherwise would have taken more than two months by hand. Rakuten said, "At the highest effort, Fable reflects on and validates its own work. For us, that's what makes highly autonomous operations possible."

Those examples explain why Anthropic is charging at the top of its public lineup. Buyers get a model Anthropic says can stay with long-running coding, knowledge-work and research tasks. They also get classifier fallbacks and a 30-day retention policy for Mythos-class traffic.

Anthropic has moved Mythos-class capability into the public market, but less-restricted Mythos 5 access remains behind Project Glasswing and the planned biology trusted-access track. GitHub administrators can still leave Fable off. After June 23, credits and the retention switch become the first measurable check on how much enterprise customers want Mythos-class work under monitoring.

Frequently Asked Questions

What is Claude Fable 5?

Claude Fable 5 is Anthropic’s public Mythos-class model. It uses the same underlying model as Mythos 5, but routes high-risk categories to Opus 4.8 through safeguards.

How does the Opus 4.8 fallback work?

Anthropic says classifiers detect cybersecurity, biology and chemistry, or distillation requests. When they trigger, the response is handled by Opus 4.8 and the user is informed.

What data retention applies to Fable 5?

Anthropic requires 30-day retention for Mythos-class traffic for safety monitoring. GitHub says Fable 5 prompts and outputs in Copilot can be retained up to 30 days.

Who can use Claude Mythos 5?

Mythos 5 is limited to Project Glasswing partners and planned trusted-access biology users. Anthropic says it will expand access in consultation with the U.S. government.

Why does Fable 5 pricing matter?

Fable 5 costs $10 per million input tokens and $50 per million output tokens. That is twice Opus 4.8, making the safety and retention tradeoffs part of the buying decision.

AI-generated summary, reviewed by an editor. More on our AI guidelines.

OpenAI Builds Cybersecurity Product for Select Partners as AI Hacking Fears Mount
OpenAI is finalizing a product with advanced cybersecurity capabilities for a limited set of partners, Axios reported Thursday, becoming the second major AI lab in a week to restrict access to tools c
The Implicator
Two Cyber Models, Two Opposite Bets. The Subsidy Era Ends.
San Francisco | Wednesday, April 15, 2026 OpenAI shipped GPT-5.4-Cyber to thousands of verified defenders on Tuesday, exactly one week after Anthropic restricted Mythos Preview to roughly forty vette
The Implicator
OpenAI Ships GPT-5.4-Cyber, Expands Trusted Access to Thousands of Defenders
OpenAI on Tuesday unveiled GPT-5.4-Cyber, a variant of its flagship model fine-tuned for defensive security work, and opened tiered access to thousands of verified defenders through its Trusted Access
The Implicator
AI News

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai