US-based Anthropic released Claude Fable 5, the first publicly available model in its most capable AI class, on June 9, two months after the company said an equivalent system was too dangerous to put in general circulation.

The launch sharpens a tension at the centre of the US AI industry, where firms racing to demonstrate ever-greater capability to investors are simultaneously warning that the technology is advancing faster than safety measures can keep pace, a contradiction made starker by Anthropic's confidential filing for a stock market listing days earlier.

Fable 5 draws on the same underlying technology as Claude Mythos, which Anthropic unveiled in April and restricted to a small group of organisations over fears it could be used to attack computer systems. The company said a public release was now possible because it had added safeguards that block responses on high-risk subjects such as cybersecurity, biology and chemistry, redirecting those queries to a less capable model, Claude Opus 4.8.

"Fable's capabilities exceed those of any model we've ever made generally available," the company said in a blog post, adding that releasing a model of that strength "comes with risks".

The safeguards trigger in under 5% of sessions and remain stricter than the company considers ideal, according to its announcement. Anthropic said an external bug bounty programme running more than 1,000 hours of testing found no universal way to bypass the model's restrictions.

Joel Pen, a member of Anthropic's technical staff, told CNBC that Fable 5 marked a significant jump in capability that required additional guardrails to prevent misuse, citing the example of a user asking how to produce the toxin ricin.

The release lands as Anthropic, led by chief executive Dario Amodei, prepares for an initial public offering after a funding round valued the firm at $965bn (£721bn), ahead of rival OpenAI. Anthropic said in May its revenue run rate had reached $47bn, up from about $10bn a year earlier.

Organisations already testing Mythos will gain access to an updated version, Claude Mythos 5, without the cybersecurity and biology limits, the company said. Firms using the earlier model have reported finding more than 10,000 critical security flaws in their systems.

Fable 5 is priced at roughly twice the level of Opus, at $10 per million input tokens and $50 per million output tokens, CNBC reported.