Anthropic recently launched Claude Fable 5, the first Mythos-class AI model designed for public use. The company claims that Fable 5 offers similar capabilities as Mythos AI while adding safety measures. While Microsoft quickly made the model available to its customers, the company is reportedly limiting its use internally. According to a report by The Verge, Microsoft has not added Claude Fable 5 to the internal version of GitHub Copilot that employees use for work. Other Claude models are available because they operate under Zero Data Retention (ZDR) rules, which do not store prompts or responses.
The reported restriction is linked to new data retention rules introduced with Claude Fable 5. Unlike previous Claude models, Fable 5 requires Anthropic to temporarily store prompts and outputs in order to operate new safety classifiers. Anthropic reportedly keeps the prompts and outputs data for up to 30 days before deleting it.
Also read: AI may bring cyberattacks and job losses if left unchecked, warns Anthropic CEO Dario Amodei
The situation becomes more complicated when Anthropic’s safety systems flag content as a potential violation of its usage policies. In such cases, prompts and outputs may be stored for as long as two years. This has reportedly raised concerns inside Microsoft about how employee data, customer information and other confidential content could be handled.
Microsoft has said to inform employees that its legal teams are currently reviewing Anthropic’s data retention requirements. As of now, it is unclear if Microsoft will eventually approve the model for employee use.
Interestingly, as soon as the Claude Fable 5 was announced, Microsoft quickly rolled out Claude Fable 5 to its GitHub Copilot and Foundry customers. So, it seems like that the tech giant is taking a more careful approach when it comes to internal adoption.
Also read: Anthropic launches Claude Fable 5, a safer Mythos AI model: Features, availability and how to use it
Claude Fable 5 arrives just weeks after Anthropic claimed that the Mythos AI is so capable at cybersecurity tasks that it was too dangerous to release publicly. As per Anthropic, Fable 5 performs better than its previous public AI models. The model can handle complex software projects, understand visual information, analyse large amounts of data, and even stay focused during long tasks involving millions of tokens.
Also read: Starlink India launch reportedly on hold over security concerns: Here is what happened