Claude’s up to date Fable and Mythos fashions watermark their replies
In its announcement, Anthropic reveals off quite a lot of benchmarks demonstrating the efficiency boosts the 2 fashions, full with double-digit will increase with regards to agentic scientific analysis, agentic coding, and enterprise workflows.
Anthropic additionally says each Fable 5.1 and Mythos 5.1 have been evaluated utilizing the corporate’s new security and alignment procedures, developed in response to Anthropic’s disclosures of Claude fashions hacking exterior corporations throughout inner testing. Particularly, Mythos 5.1 is “much less possible than Mythos 5 to attempt to entry assets exterior of its check setting when assigned an in any other case unimaginable process,” Anthropic stated.
Past their agentic capabilities and upgraded safeguards, via, Claude Fable 5.1 and Mythos 5.1 mark the primary Claude fashions to watermark their textual content and file outputs, in accordance with the EU AI Act.
“As we lately defined, this watermark is invisible to anybody who doesn’t have the detection API,” Anthropic wrote in its announcement. “It has no sensible influence on the standard or content material of Claude’s outputs and incorporates no details about the person, their group, or their conversations with Claude.”
As for the detection API, Anthropic says is now accessible to regulators, regulation enforcement, media, fact-checkers, impartial researches, and different “eligible teams” as outlined by the just-enacted EU regulation, including that it’ll “develop entry to the detection API over time.” You may apply for entry to the API by way of this signup kind.
Anthropic had beforehand said it might add watermark assist to any new Claude fashions launched after August 2, 2026. Claude Fable 5.1 and Mythos 5.1 are the primary new Claude fashions to reach subsequent to that date, whereas older Claude fashions will ultimately add watermarking assist as properly, Anthropic has beforehand stated.
