Anthropic releases AI model it previously said was too powerful to make public

Anthropic releases AI model it previously called too risky for public access

Anthropic released Claude Fable 5 to the public on Tuesday, a version of its Claude Mythos AI system that company executives had said posed security risks if made widely available. The model comes with safeguards and user limits, though Anthropic acknowledged that "releasing a model this capable comes with risks."

Mythos was first shared with a limited group of organizations in April for testing. Anthropic restricted access then because the tool's ability to identify and exploit computer system vulnerabilities could cause harm if widely deployed.

About 150 organizations that tested Mythos will now gain access to Claude Mythos 5, which removes restrictions on cybersecurity and biology applications for approved users. Anthropic said this access remains limited to "a small group of cyberdefenders and infrastructure providers" but plans to expand it.

Government agencies are already testing the model

U.S. government agencies have been testing Mythos despite Anthropic's ongoing lawsuit against the Department of Defense over government use of its tools. Canadian Finance Minister François-Philippe Champagne told the BBC in April that concerns about Mythos were justified because "it's the unknown, unknown."

Organizations testing Mythos have reported finding more than 10,000 critical security flaws in their systems using the tool.

Questions about timing and business incentives

Anthropic's public release comes as the company approaches a valuation near $1 trillion and prepares to go public. Demonstrating advanced capabilities typically strengthens investor confidence in AI companies.

Some observers have questioned whether concerns about Mythos's risks reflect genuine dangers or marketing strategy. The company's framing of the model as exceptionally powerful could benefit its market position.

Anthropic co-founder Jack Clark said last week that AI development is advancing so quickly the industry needs a brake. "Right now, it's like the AI industry has a gas pedal, but it doesn't have a brake pedal," he told BBC Newsnight.

Both Fable and Mythos can work unattended on tasks for longer periods than previous Claude versions, Anthropic said.

For government professionals evaluating these tools, resources on AI for Government and Claude may provide relevant context.