Anthropic Releases Claude Fable 5 as Its First Public Mythos-Class Model

Anthropic has released Claude Fable 5, its first generally available Mythos-class AI model, with built-in safeguards that route some high-risk requests to Claude Opus 4.8 instead.

The company announced Claude Fable 5 and Claude Mythos 5 on June 9, 2026. Fable 5 is based on the same underlying model as Mythos 5, but Anthropic added restrictions for areas such as cybersecurity, biology, chemistry, and model distillation.

Anthropic says Fable 5 sits above the Opus line in capability and is built for long, complex work. The release also brings new debate over how AI companies should balance powerful public models, safety controls, transparency, and enterprise data policies.

Claude Fable 5 brings Mythos-level capabilities to more users

On the official Claude Fable 5 page, Anthropic describes the model as designed for ambitious, long-running projects, including multi-stage coding tasks, enterprise workflows, document-heavy analysis, and vision work.

The company says Fable 5 is its most capable generally available model and performs strongly on coding, knowledge work, vision, and computer-use tasks. Anthropic says the model can also work in agent harnesses such as Claude Code or Claude Managed Agents for extended projects.

Pricing starts at $10 per million input tokens and $50 per million output tokens. Anthropic also says Fable 5 is available through the Claude API, consumption-based Enterprise plans, supported marketplaces, Amazon Web Services, Google Cloud, and Microsoft Foundry.

Model	Claude Fable 5
Model tier	Mythos class
Availability	Generally available, with staged subscription access
Best use cases	Long coding projects, agentic work, analytics, knowledge work, and vision tasks
Input price	$10 per million tokens
Output price	$50 per million tokens
Safety fallback	Some sensitive requests route to Claude Opus 4.8
Data retention	30 days for Mythos-class traffic used for safety monitoring

Why Anthropic added safeguards to Fable 5

Anthropic says Mythos-class models are powerful enough to create new risks in dual-use areas. The company says the same capabilities that can help defenders find security flaws could also help malicious users with offensive cyber tasks.

The Claude Mythos 5 page says Mythos 5 is available only to a small set of initial testing partners for cybersecurity, with biology research access planned through trusted programs.

Fable 5 is the broader public version of that capability. When its classifiers detect requests related to cybersecurity, biology and chemistry, or distillation, the response is handled by Claude Opus 4.8 instead of the full Fable model.

• Cybersecurity safeguards cover exploitation and broader offensive cyber tasks.
• Biology and chemistry safeguards are tuned broadly to reduce misuse risk.
• Distillation safeguards target attempts to use Fable 5 to train competing models.
• Anthropic says most sessions do not trigger fallback.
• Users are supposed to be informed when fallback occurs.

Claude Mythos 5 remains restricted to trusted users

Anthropic is also launching Claude Mythos 5 for a restricted group of cyber defenders and infrastructure providers. It uses the same underlying model as Fable 5, but with safeguards lifted in selected areas for approved users.

The company says Mythos 5 will initially be deployed through Project Glasswing, a program created to give selected cybersecurity partners access to advanced models for defensive work.

Anthropic says it plans to expand trusted access over time. It also says a future biology access program will let selected researchers use Mythos-class capabilities with biology and chemistry safeguards removed, while keeping cyber safeguards in place.

Feature	Claude Fable 5	Claude Mythos 5
Underlying model	Same base model	Same base model
Access	Broadly available	Restricted trusted access
Cyber safeguards	Enabled	Lifted for approved cyber partners
Biology and chemistry safeguards	Enabled	Planned removal for selected biology researchers
Primary audience	Developers, enterprises, analysts, and general advanced users	Cyber defenders, infrastructure providers, and approved researchers

Fable 5 uses Opus 4.8 fallback for sensitive prompts

Anthropic says its safeguards trigger in less than 5% of sessions on average. For the remaining sessions, Fable 5 should deliver the same performance as Mythos 5.

According to TechCrunch, Anthropic said Fable 5 falls back to Claude Opus 4.8 in high-risk areas such as cybersecurity, biology, chemistry, and distillation, while still giving users access to Mythos-level capabilities for most other work.

This approach is meant to avoid a full refusal when a safer model can answer. However, it also means some users may not always receive a response from the model they selected, especially when working near sensitive topics.

The launch sparked concern over transparency

After launch, Anthropic faced criticism over a safeguard for AI development and distillation-related prompts that could change or degrade responses without clearly telling users. That led to concerns from developers and AI researchers.

The Verge reported that Anthropic apologized for the invisible guardrail approach and said it would make the safeguard visible, with flagged requests falling back to Claude Opus 4.8 and users being notified.

The change matters because Fable 5 is aimed at serious technical work. Researchers and developers need to know when a model’s answer reflects the selected model and when a safety layer has changed the response path.

• Anthropic initially used invisible safeguards for some distillation-related cases.
• Developers criticized the lack of user visibility.
• Anthropic said it made the wrong tradeoff.
• The company moved toward visible fallback to Opus 4.8.
• The issue highlights growing tension between AI safety and model transparency.

Data retention becomes part of the safety tradeoff

Anthropic is requiring 30-day data retention for Fable 5, Mythos 5, and future models with similar or higher capability levels. The company says it will use the retained data for safety monitoring, not to train new Claude models.

The Anthropic launch post says the retained traffic helps detect complex attacks, novel jailbreaks, multi-request abuse, and false positives in safeguards.

That policy may create friction for enterprises that prefer zero-retention arrangements. It also explains why Fable 5 is likely to receive closer review from legal, security, and compliance teams before broad internal deployment.

Project Glasswing shaped the Mythos release

Anthropic first introduced Mythos-class access through Project Glasswing, which focuses on helping cyber defenders use advanced AI to secure important software and infrastructure.

The company says Mythos Preview showed how advanced models can help defenders find and fix vulnerabilities. At the same time, Anthropic warned that similar capabilities could make cyberattacks cheaper and more frequent if released without safeguards.

Fable 5 reflects that compromise. Anthropic is releasing more capability to the public while keeping the highest-risk cyber and biology use cases behind fallback systems or trusted access programs.

What users can do with Claude Fable 5

For most users, Fable 5 is positioned as a high-end model for long and difficult tasks rather than quick everyday chat. Anthropic highlights coding, code migration, research, analytics, document work, visual reasoning, and agentic workflows.

The Fable product page says the model can plan across stages, delegate to sub-agents, test its own work, and evaluate visual output against a target design or goal.

That makes it most relevant for teams that need help with complex software projects, deep analysis, legal and finance workflows, product prototypes, and other tasks where a cheaper model may need too much supervision.

Use case	Why Fable 5 may help
Large code migrations	Can work across long, multi-step engineering tasks
Agentic coding	Can plan, test, and revise work over longer sessions
Enterprise analysis	Can process complex documents, tables, charts, and files
Vision tasks	Can interpret diagrams, screenshots, charts, and nested visual content
Research workflows	Can help with long-form reasoning and structured synthesis

What to watch next

The biggest questions now concern reliability, cost, and trust. Fable 5 is more expensive than lower-tier Claude models, and its safety fallback can affect sensitive technical or scientific workflows.

The Mythos page shows that Anthropic sees these models as powerful enough to require trusted access programs for some capabilities. That approach may influence how other AI companies release frontier models.

TechCrunch also noted that Anthropic’s mandatory 30-day retention policy could become a new industry precedent for very capable models, especially when companies frame monitoring as a safety requirement.

For now, Claude Fable 5 gives public users access to a more capable Claude model, but not without limits. The launch shows how the next generation of AI products may come with more power, higher prices, stronger safeguards, and more visible debate over who gets unrestricted access.

The latest criticism covered by The Verge also shows that users expect transparency when model behavior changes because of hidden or visible safety systems. Anthropic’s response may shape how future frontier AI launches explain safeguards from day one.

FAQ