Anthropic releases Fable 5, its most powerful AI, with guardrails on cybersecurity and biological risks
The American start-up made its most advanced model available to the public on Tuesday, while keeping a more powerful, unrestricted version for vetted enterprises and government agencies.
Anthropic released Fable 5 to the general public on Tuesday, marking the first time a model from its most advanced Mythos class has been made widely accessible. The launch pairs a consumer-facing version with deliberate restrictions in sensitive areas with a separate, unrestricted model for trusted partners.
A two-tier release
Fable 5 is the secured, public version of the Mythos family, which Anthropic first disclosed in April but had kept restricted for safety reasons. In parallel, the company is offering Claude Mythos 5, an unrestricted version, to enterprises, organisations, and state agencies that already have access to this model family. Anthropic describes the unrestricted model as capable of detecting and exploiting security vulnerabilities at unprecedented speed and precision.
à une vitesse et avec une acuité inédites
Expanded safety boundaries
Anthropic has broadened its safety restrictions beyond cybersecurity to include biological and chemical risks. The company cited the example of designing adeno-associated viruses (AAV), which can be dangerous but also beneficial in gene therapy. For most queries touching on cybersecurity or biology, the system defaults to responses from the less powerful Opus 4.8 model, which was made public at the end of May. Selected biology researchers will receive access to the unrestricted version.
Political and institutional friction
The April announcement of Mythos drew accusations of fear-mongering marketing and triggered visible agitation among governments and global institutions assessing threats to critical infrastructure, including banking and energy systems. Several enterprises that gained access to Mythos have since confirmed its capabilities. The US government, despite ongoing disputes with Anthropic, eventually tested the model. The White House subsequently introduced an optional evaluation framework for the most powerful American AI models before their commercial release.
Tensions with the Pentagon
Anthropic's stance on safety has placed it in an unprecedented standoff with the Trump administration. The company refused to lift its restrictions on mass surveillance and lethal autonomous weapons. In response, the Pentagon terminated its contracts with Anthropic, whose AI tools had been the only ones cleared for classified defence work.
Competitive landscape
The launch arrives as OpenAI, Anthropic's chief rival, prepares to unveil version 5.6 of ChatGPT. The broader AI sector is watching how the dual-track release strategy — a restricted public model alongside a high-capability enterprise version — shapes both market dynamics and regulatory conversations.
- Anthropic announces the existence of the Mythos model class and restricts access to trusted partners for cyber protection.
- Opus 4.8, a less powerful model, is made public and later used as a fallback for sensitive queries.
- Fable 5, the first public Mythos-class model, launches with restrictions; Claude Mythos 5 offered to vetted partners.
