Since Anthropic unveiled the Mythos model in April 2026 to a restricted group of organizations through Project Glasswing, the company has been working toward a broader release it always called an "eventual goal." That day arrived Tuesday, June 9.

Anthropic released Claude Fable 5 — its first Mythos-class model available to the general public. Per TechCrunch, Fable 5 is now accessible through Anthropic's Claude API and consumption-based Enterprise plans, with staged rollout to Pro, Max, Team, and seat-based Enterprise subscribers.

What Mythos-Class Models Mean

Mythos-class models represent a significant capability jump. When Anthropic first unveiled Mythos in April, it refused to make it publicly available specifically because of what the model could do with cybersecurity tasks. The company was essentially saying: this thing is powerful enough that we don't trust the world with it yet.

That threshold has now been crossed — with conditions.

According to CNBC, Fable 5 scores more than 10% higher than Claude Opus 4.8 on some benchmarks. Anthropic's head of product management for research, Dianne Penn, told CNBC the model represents a "significant jump" in capability. The Verge reports that Anthropic claims Fable 5 shows "exceptional performance in software engineering, knowledge work, and vision," with its advantage over competitors growing as tasks get longer and more complex.

The Safety Architecture — And Its Limits

Anthropic built in hard blocks for high-risk areas: cybersecurity, biology, chemistry, and distillation, according to TechCrunch. When a query trips one of those wires, the model falls back to Claude Opus 4.8. Anthropic says 95 percent of Fable sessions run entirely on Fable responses without hitting that fallback.

The company stress-tested the safeguards before release. Per TechCrunch, Anthropic ran an internal bug bounty producing no universal jailbreaks across over 1,000 hours of testing. External red-teaming organizations also came up empty. That's also not a guarantee — novel jailbreaks happen, and 1,000 hours of controlled testing is different from millions of users worldwide probing the system 24/7.

Anthropic itself acknowledges this gap. Which is why it added something that deserves significant attention in the current coverage.

Mandatory Data Retention — The Policy Shift

With the Fable 5 and Mythos 5 launch, Anthropic is requiring a 30-day retention on ALL traffic — including for enterprises that previously had zero-retention agreements. According to TechCrunch, Anthropic says it won't use the data for training and will only use it to defend against novel jailbreaks and reduce false positives.

This is a significant policy shift. Enterprise customers signed contracts with specific privacy terms. Those terms just changed, unilaterally, because Anthropic decided its safety needs override pre-existing zero-retention agreements.

Anthropic frames this as a safety measure. But it also sets a precedent: the more powerful the model, the less privacy you get. TechCrunch flags this directly, noting the policy "could set an industry precedent." If this becomes the norm, enterprises feeding sensitive data into AI systems will have to accept mandatory retention of their queries as the cost of access to frontier models. The question deserves public debate, not a footnote in a product launch blog.

The Pricing Rollout Timeline

The rollout schedule is also worth scrutinizing.

Per TechCrunch, Fable 5 will be included in Pro, Max, Team, and Enterprise plans at no extra cost through June 22. On June 23, Anthropic pulls it from subscription plans and requires usage credits. The company says it plans to restore Fable 5 as a standard subscription feature "as soon as possible" — with no defined timeline.

That's a two-week free window followed by a pivot to consumption-based billing. The Verge notes pricing for both Fable 5 and Mythos 5 is "significantly higher" than Anthropic's former flagship. For businesses evaluating adoption, the June 22 deadline creates artificial urgency — a classic enterprise sales tactic dressed up as a staged rollout.

Mythos 5: The Version Nobody Gets to See

Alongside Fable 5, Anthropic is releasing Claude Mythos 5 — the same underlying model as Fable 5, but with some safeguards lifted. According to The Verge, access is currently limited to organizations already approved through Project Glasswing, with plans to expand through a "more systematic trusted-access program" over time.

Anthropic did not respond on the record to The Verge's questions about how Mythos 5 relates to the original Mythos Preview, or why the models are numbered "5" when no Fable 1 through 4 or Mythos 1 through 4 were ever released. Both questions deserve answers.

The Safety Debate

Critics — including some AI safety researchers — argue that releasing any version of a model previously deemed too dangerous, regardless of safeguards, normalizes a pattern: build something risky, add a filter, call it safe, ship it. The concern isn't that Fable 5 will immediately enable mass harm. It's that each successive release moves what counts as "safe enough," and that competitive pressure — including an anticipated IPO flagged by both TechCrunch and CNBC — is driving the timeline.

Anthropic is a company preparing to go public. Anthropic also previously warned, publicly, that AI systems may soon achieve recursive self-improvement. Both things are true simultaneously. Investors want revenue. Revenue requires product releases. The question of whether this release schedule is primarily safety-driven or market-driven is legitimate.

Anthropic has been more transparent about its safety process than most competitors. The fallback architecture, the bug bounty data, and the limited Mythos rollout before broad release are all more rigorous than OpenAI's typical approach.

What This Means for You

If you're an enterprise customer, read your new contract terms before June 23. The zero-retention deal you signed may no longer exist.

If you're a developer, you have two weeks to access the most powerful publicly available AI model at no extra charge.

If you're a taxpayer watching the broader AI race unfold — China is spending $297 billion on data centers, Apollo and Blackstone just put $35 billion behind Anthropic's chip expansion, and the most capable AI model the company has ever publicly shipped dropped today. The stakes are real. The window is narrow. And the safety questions don't have clean answers yet.