Anthropic is finally letting the public near an AI system once considered too dangerous to release—just not the original version. The company on Tuesday rolled out Claude Fable 5, a heavily constrained spin on its Mythos system, which it previously warned could become a potent hacking aid, the New York Times reports. Fable leans on older tech for some tasks and adds strict filters meant to block help with cybersecurity exploits, biology, and other sensitive areas. That means it should be harder for criminals to use it to break into networks—but also harder for defenders to fully lean on it to secure them.
Risky queries will largely be routed to another Anthropic model, Claude Opus 4.8, which was designed to avoid Mythos-style dangers and launched last month. The original Mythos was available only to a select set of critical infrastructure operators, a move that split security experts: some praise the caution, others argue broader access would speed defensive research.
- In a blog post, Anthropic said it is releasing Mythos 5 along with Fable 5. Like the earlier version, it is available only to a small group of "cyberdefenders and infrastructure providers," but the company plans to bring in a "broader trusted access program." The model, Anthropic said, "has the strongest cybersecurity capabilities of any model in the world."
Benchmark firm Vals AI says Fable edges out all other public AI systems overall and is especially strong at coding and math, though it's weaker in areas like health care and taxes—and costs twice as much as Opus. Anthropic says the safety work is ongoing, not solved. "What we wanted to do was to be very intentional about building new types of classifiers and new types of safety guardrails in place for this launch," Dianne Penn, Anthropic's head of product management for research, tells CNBC. The aim, she says, is to provide the technology with "the right safety guardrails so that it can do asymmetrically more benefits than harm."
,
uBlock 
,
and
Ghostery 
)
