Anthropic, the San Francisco‑based AI firm behind the Claude series, announced today that its most capable model—formerly known only as the Mythos class—will be accessible to customers as Claude Fable 5. The rollout appears on the company’s Pro, Max, Team and seat‑based Enterprise subscription tiers, with no extra fee for a limited period.

For months, Anthropic described the Mythos model as "too dangerous to release," citing concerns that its advanced capabilities could be weaponized for cyber‑attacks, the creation of harmful biological agents, or the acceleration of AI development beyond current safeguards. The company’s messaging emphasized the need for "extraordinary safeguards" before any public deployment.

Fable 5 marks a sharp pivot. Anthropic says the model now includes extensive guardrails, such as separate AI classifiers that monitor requests for signs of malicious intent. When a request triggers a cybersecurity flag, the system hands the query off to Claude Opus 4.8, keeping the most sensitive functions behind an additional layer of protection.

Beyond safety, the firm touts Fable 5’s ability to sustain work over hours, days, or even longer. Unlike earlier Claude versions that respond to isolated prompts, the new model can retain context across lengthy projects, making it suitable for complex software engineering tasks, in‑depth research, and multi‑step AI agent workflows. Anthropic frames the model as a "digital collaborator" rather than a simple conversational assistant.

Early adopters have responded positively to the performance boost. A Reddit user noted, "Fable 5 is insanely good but watch your usage; I was burning 2% a minute on 20x." Another commented that the model’s refusal to handle certain cybersecurity queries highlights a "preview of AI inequality," where the public receives a sanitized version while trusted institutions may retain fuller access.

Access to Fable 5 will shift on June 23 to a usage‑credit system intended to manage demand rather than impose a permanent paywall. Anthropic explained that once capacity allows, the model will return to a standard part of subscription plans. The temporary change aims to balance the high token consumption reported by power users with the need to keep the service reliable for all subscribers.

Anthropic’s decision to open the Mythos‑class capabilities, even with added safeguards, underscores a broader industry debate about how much advanced AI should be democratized. The company claims the model possesses "the strongest cybersecurity capabilities of any model in the world," a double‑edged sword that could aid security researchers while also offering a potent tool for malicious actors.

While the rollout has been met with enthusiasm for its technical prowess, the conversation around risk mitigation continues. Anthropic’s layered approach—pairing the model with classifiers and a fallback system—will be tested in real‑world usage. Observers will watch whether the safeguards hold up as more developers, researchers and enterprises integrate Fable 5 into their workflows.

For now, Claude Fable 5 represents a significant step in Anthropic’s strategy to move AI from a conversational assistant toward a more autonomous, collaborative partner, all while navigating the tightrope of safety and accessibility.

Cet article a été rédigé avec l'assistance de l'IA.
News Factory APP - actualités agentiques pour booster votre SEO et AEO.