The artificial intelligence company Anthropic has released what it describes as its most capable publicly available system to date, giving ordinary users their first access to a class of technology that earlier this year alarmed cybersecurity experts and government officials.
The new model, called Fable 5, was made available on Tuesday afternoon. According to NBC News, it is the first product released to the general public from the same family as Anthropic’s powerful Mythos systems, which drew attention in April for what the company has called a superhuman ability to find and exploit weaknesses in computer software. In a blog post announcing the launch, the firm said Fable’s abilities “exceed those of any model we’ve ever made generally available”, describing it as state of the art across almost every benchmark it had been tested against, with particularly strong performance in software engineering, scientific research, vision and general knowledge work.
The performance figures cited by the company are striking. As reported by Fast Company, Fable 5 scored 80.3 per cent on SWE-Bench-Pro, a test of an AI model’s coding ability, well ahead of the 58.6 per cent attributed to OpenAI’s GPT-5.5 and the 54.2 per cent recorded by Google DeepMind’s Gemini 3.1 Pro. The analytics firm Hex, which assesses long and complex analytical tasks, said Fable 5 was the first model to pass 90 per cent on its core benchmark.
What makes the release notable is not only the model’s power but the way Anthropic has chosen to contain it. The company says Fable 5 draws on the same tier of technology as Mythos, an earlier version of which was able to identify thousands of serious vulnerabilities across major operating systems and web browsers. To make such capability safe for wider use, Anthropic has built in guardrails designed to block responses to questions touching on sensitive subjects, chiefly cybersecurity and biology. When a user raises one of those topics, the query is instead routed to an older and less powerful model, Opus 4.8, which until Tuesday had been the firm’s best publicly available system.
The reasoning behind that approach reflects long-standing anxieties within the field. Researchers have warned that increasingly capable AI could assist criminals in attacking banks, power grids or other critical infrastructure, while others fear such systems might one day help in the design of biological weapons. “The same queries that are beneficial in the hands of cybersecurity professionals and biology researchers could be dangerous if available to malicious actors,” the company wrote.
Asked about the restrictions by NBC News, Anthropic said its caution was deliberate. “Fable 5’s capabilities in areas like cybersecurity, biology and chemistry are advanced enough that we’re taking a deliberately conservative approach for these topics at launch,” it said, adding that it had opted to “err on the side of caution” by applying broad limits for now. The company says the diversion to Opus 4.8 affects only around five per cent of sessions, allowing most users to receive useful answers while the riskiest questions are handled by a weaker system. Dianne Penn, the firm’s head of product management for research, framed the strategy to CNBC as part of what Anthropic calls a “race to the top”, aiming to deliver the technology in a way that does “asymmetrically more benefits than harm”.
Anthropic also said it had tried to break its own safeguards before release. Its cybersecurity team spent more than 1,000 hours attempting to find a way around the restrictions, according to Inc., and reported that it had been unable to identify a single workable method.
For users, access will initially be free. The company said Fable 5 would be available at no extra cost to subscribers on its Pro, Max, Team and seat-based Enterprise plans until 22 June, after which it can be reached only by purchasing additional computing credits. Anthropic said it hoped to fold the model into its standard plans “as quickly as we can”, though it acknowledged the strain on its resources; the firm has struggled recently to meet soaring demand, and the new model is likely to add further pressure on its limited computing capacity.
Alongside the public launch, Anthropic announced an upgraded system, Mythos 5, for the trusted partners who had already been granted access to the original Mythos Preview. That earlier model had been shared with more than 150 organisations worldwide — among them financial institutions, software firms and healthcare networks — to help them strengthen their defences ahead of any broader release. The company said Mythos 5 had set new performance records in fields including drug design and molecular biology, and described it as its first model to consistently generate what it called novel and compelling scientific hypotheses.
The timing of the release is significant. As TechCrunch noted, it arrives as Anthropic prepares to enter the public markets alongside OpenAI and Elon Musk’s SpaceX, and shortly after the company publicly urged the world’s leading AI laboratories to agree a coordinated check on frontier development. It also follows by a week an executive order signed by President Donald Trump, which seeks to create a voluntary system for AI companies to share their technology with the government for safety testing before public release, and to strengthen the state’s own cyber defences.
