Microsoft Limits employees’ use of Anthropic‘s Cloud Fable 5 While its legal teams evaluate changes to Anthropic’s data retention requirements, Reuters reported I mentioned Wednesday (June 10), citing a paywall condition By the edge.
Anthropic’s data retention policy for the newly released Claude Fable 5, a Mythos-class model, includes retaining claims and outputs for 30 days on each platform on which the model is submitted, and for up to two years if claims and outputs are flagged by Anthropic’s trust and safety classifiers as violating the company’s usage policy, according to the report.
The report said that the company retains this data for trust and safety purposes.
Anthropic announced on Tuesday (June 9) that it has released Claude Fable 5 after development Guarantees To prevent its misuse and other Mythos-class models.
When the company announced its first Mythos model, the Claude Mythos Preview, in April, it limited the model’s release to select partners so they could use its features. Cyber security capabilities to enhance its systems.
Anthropic said Tuesday that Claude Fable 5 is safe for general use, with safeguards designed to prevent its misuse for cybersecurity, biology, chemistry or distillation.
Advertisement: Scroll to continue
The company added that the model has capabilities beyond those of any previous model made publicly available by Anthropic, is state-of-the-art in most AI capability standards, and performs particularly well in software engineering, knowledge work, vision, and scientific research.
On Tuesday advertisement Following the model’s release, Anthropic said it had introduced a new data retention policy for Claude Fable 5 and other models with higher levels of similar capability.
Anthropic said the new 30-day data retention policy applies to both first-party and third-party surfaces and that the company will ensure data is deleted after 30 days in “almost all cases.” The company added that it will not use this data to train new Cloud models, or for any other purpose not related to safety, and that it records all human access to the data.
“The data will help us defend against sophisticated and new attacks (including new jailbreaks and attacks that operate across multiple requests) as well as help us identify and reduce false positives,” Anthropic said in its announcement.
