Anthropic opens Mythos-class AI to the public with data retention
Anthropic launched Claude Fable 5 for public use and Claude Mythos 5 for vetted cyber partners, while mandating 30-day data retention for both.
Intelligence analysis by GPT-5.4 Mini
Anthropic has made its most capable Mythos-class model broadly available as Claude Fable 5, but only with guardrails that divert risky requests to a weaker model. At the same time, it tightened access and monitoring by requiring 30-day retention of traffic across both public and restricted versions.
Anthropic has opened a very powerful robot helper to everyone, but it puts a guard at the door and stores the chats for 30 days. It is like a fancy library where some books are easy to borrow, but the rare ones are only for trusted people.
Analysis
What Anthropic launched
Anthropic introduced two Mythos-class models on June 9, 2026. Claude Fable 5 is the public version, while Claude Mythos 5 is reserved for vetted cybersecurity defenders and critical-infrastructure partners. The company says both use the same underlying model, with identical weights, capabilities, and pricing, but access rules differ sharply.
How the split works
Fable 5 sits behind classifiers that screen prompts before they reach the model. When a request touches sensitive areas such as cybersecurity, biology, chemistry, or model distillation, the system routes the request to Claude Opus 4.8 instead of answering directly. Anthropic says more than 95% of sessions do not trigger fallback, so most users should see the full experience. MediaNama’s own test found Fable 5 routing a routine open-source web app audit to Opus 4.8.
Mythos 5, by contrast, is available only to Project Glasswing cyber defenders and select infrastructure partners, with the cybersecurity safeguards lifted. Access depends on the user’s role, not on a paid upgrade.
The retention policy
Anthropic also imposed a mandatory 30-day retention rule on traffic from both models, including enterprise use cases that previously had zero-retention agreements. The company says it will use the data only to defend against novel attacks and reduce false positives, not to train models. It also says human access to the data will be logged and the data will usually be deleted after 30 days.
India angle
The article says India was excluded from the restricted-access program until April, before Anthropic expanded access to more than 15 countries and brought in a small number of Indian entities from government and private sectors. Their identities have not been disclosed. That leaves the public Indian market with guardrailed access to Fable 5, while Mythos 5 remains limited to a narrow, unnamed set of users.
Why this release matters
This is a public launch of Anthropic’s strongest model, but it comes with tighter control, more logging, and a clear warning that the company sees safety as something to enforce centrally. That makes the product announcement also a policy signal for how frontier AI may be offered to enterprises in India and elsewhere.
Key points
- Anthropic launched Claude Fable 5 for the public and Claude Mythos 5 for vetted cybersecurity partners.
- Both versions use the same underlying model, but Fable 5 adds classifiers that divert risky prompts to Claude Opus 4.8.
- Anthropic requires 30-day retention of traffic on both models, even for some enterprises that previously had zero-retention deals.
- India now has limited public access to the guarded version, while Mythos 5 remains restricted to a small, unnamed group of users.
- Anthropic says the retained data will be used for security purposes, not for model training.
If the safeguards work as intended, more users can access stronger AI without exposing the most dangerous capabilities as openly. Anthropic’s retention policy may also help it spot novel attacks and cut false alarms, which could make the system safer to use at scale.
The mandatory retention rule means customers who expected zero-retention handling lose that privacy posture, which could be a concern for Indian enterprises. The article also suggests the public release expands both defensive and offensive potential, so wider access could increase misuse if the filters miss risky prompts.
