The Rundown AI is reporting that Anthropic's Fable 5 artificial intelligence model is back online after being pulled offline by the U.S. government on June 12. The model's return, nearly three weeks later, includes new guardrails and a commitment for the U.S. government to have pre-release access to Anthropic's future models. The Commerce Department lifted its export controls, which were initially prompted by Amazon researchers identifying security flaws in Fable's outputs. The Rundown AI said Fable 5 is now available across Claude tiers and platforms, with paid plans having usage capped until July 7. An updated safety filter blocks the identified cybersecurity issue more than 99% of the time, though Anthropic warned it might occasionally flag harmless coding requests.
Meta is developing a cloud service to rent out its spare data center capacity, a move that boosted the company's stock by 9.3%. The Rundown AI said options for this service could include renting compute power or allowing developers to access Meta-hosted models, such as its recently launched Muse Spark. CEO Mark Zuckerberg previously stated that outside companies frequently inquire about purchasing Meta's compute resources, though the company still anticipates using it internally. SpaceX established a precedent in this area with xAI’s Colossus, and Anthropic, Google and Reflection AI have all signed leases this year to utilize its compute capacity.
In other AI developments, The Rundown AI highlighted a guide on using Google's Design.md tools with Claude Code to create better websites. Design.md is an open-source standard developed by Google to facilitate agent-friendly design briefs. The guide provides step-by-step instructions for setting up a project and generating website prototypes.
The Center for AI Safety and Scale Labs released its Remote Labor Index, a benchmark testing AI agents on freelance tasks graded by humans. The Rundown AI reported that Anthropic's Fable 5 achieved the highest score ever, with frontier rates climbing sixfold in a year. Fable 5 matched or surpassed human professionals on 16.1% of projects, while Opus 4.8 scored 8.3% and OpenAI's GPT-5.5 achieved 6.3%. The Rundown AI noted that despite the significant jump in performance, only about one in six freelance tasks reached professional quality by the top model, suggesting continued human involvement for judgment and refinement.