Anthropic's newly released Claude Fable 5 collects user data by default, and there is no way to turn it off. The policy, which affects all users of the model regardless of account type, has sparked backlash from privacy-conscious users, enterprise customers who previously relied on zero-data-retention agreements, and AI researchers who say the change was implemented without adequate notice.
What the Policy Actually Says
According to Anthropic's updated terms, interactions with Claude Fable 5 are logged and may be used to improve the model. This applies across API access, Claude.ai, and third-party integrations built on the model. Users who had previously negotiated zero-data-retention arrangements under older tiers are finding those protections no longer apply now that Fable 5 ends zero-data-retention for enterprise customers, marking a significant shift from earlier commitments Anthropic had made to business clients.
Key Facts
- Data collection in Claude Fable 5 is mandatory for all users with no opt-out mechanism
- Enterprise zero-data-retention agreements are no longer honored under the new model tier
- AI researchers and developers report capabilities have been quietly restricted without public changelog entries
- Anthropic released Fable 5 days after its own executives publicly warned about AI risks
- Fortune reports accusations of "secret sabotage" from developers who noticed undisclosed limitations
The timing has not gone unnoticed. Anthropic released Claude Fable amid its own AI safety warnings, with company leadership publicly flagging concerns about AI becoming too dangerous in the same week the model shipped. Critics argue the mandatory data collection policy contradicts the safety-focused image Anthropic has cultivated, since harvesting broad interaction data without consent creates its own category of risk.
"This is a fundamental change to the trust relationship between Anthropic and its enterprise users. You cannot call yourself a safety-first company and then remove the protections that safety-conscious customers specifically paid for."Enterprise developer, via Fortune
Researchers Say Capabilities Were Quietly Capped
The data collection controversy is not the only complaint circulating. Fortune reports that Anthropic is facing accusations of "secret sabotage" after developers noticed that Claude Fable 5 silently limits certain capabilities for AI researchers and power users. No public changelog or announcement accompanied the restrictions, leading to frustration among people who had built workflows around the model's previous behavior. Anthropic has not issued a detailed public response to the capability-limiting allegations as of this writing.
The restrictions appear targeted at high-volume API users and researchers probing the model's reasoning boundaries, according to developer reports. Some speculate the limits are tied to safety mitigations, while others believe they are capacity-management decisions dressed up as policy. Either way, the lack of transparency has amplified distrust at a moment when Anthropic can least afford it.
Where This Leaves Users
For everyday Claude.ai subscribers, the practical impact of mandatory data logging may be limited. Anthropic does state that data is handled according to its privacy policy and is not sold to third parties. But for legal, medical, and financial professionals who chose Claude specifically because of strict data handling promises, the policy change creates real compliance headaches.
The broader picture is one of a company navigating tension between the commercial need to collect training data and the privacy expectations it set with early adopters. As Anthropic launches Claude Fable 5 and Claude Mythos 5 as a paired release, the data policy applies to Fable 5 specifically, and it is worth noting that Mythos 5 carries different access and usage terms that have not attracted the same level of criticism so far.
Whether Anthropic walks back any portion of the mandatory collection policy remains to be seen. The company has historically been responsive to enterprise pressure, and with several large customers reportedly reviewing their contracts, a revised tiered approach to data retention could emerge. For now, users have a clear choice: accept the terms or find an alternative. There is no middle ground built into the current product.
