Anthropic releases ‘safe’ version of Claude Mythos AI model to public

2026-06-09 · Source: AI (artificial intelligence) | The Guardian · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy, Emerging Technologies & Innovation · Depth: Novice, short

Summary

Anthropic has released Fable 5, a new public version of its advanced Mythos AI model, previously restricted due to cybersecurity concerns. This model, the first from the Mythos class to be widely available, is designed for tasks like software coding, debugging, complex research questions, and image analysis. An unrestricted version, Claude Mythos 5, is offered to Project Glasswing partners, a group expanded to over 200 organizations across 15+ countries. Anthropic implemented restrictions on Fable 5 by routing sensitive cybersecurity, biology, and chemistry queries, as well as attempts to extract its technology, to the less capable Opus 4.8 model. The company also conducted over 1,000 hours of red-teaming and a bug bounty program to test these safeguards. Fable 5 is priced at \$10 per million input tokens and \$50 per million output tokens, double Opus 4.8, reflecting Anthropic's high computing costs, including a \$1.25 billion monthly datacenter lease from xAI.

Key takeaway

For Directors of AI/ML evaluating new model deployments, Anthropic's Fable 5 release highlights the necessity of balancing advanced capabilities with robust security protocols. You should consider implementing tiered access strategies and rigorous red-teaming for powerful AI models, especially those with potential cybersecurity implications. Be prepared for higher operational costs associated with safer, more capable models, as Fable 5's pricing reflects significant infrastructure investment.

Key insights

Anthropic's Fable 5 release balances advanced AI capabilities with stringent security measures and tiered access.

Principles

• Advanced AI models require tiered access for safety.
• Red-teaming is crucial for validating AI safety measures.
• Cybersecurity risks necessitate model restriction strategies.

Method

Anthropic routes sensitive queries (cybersecurity, biology, chemistry) and extraction attempts to a less capable model, Opus 4.8, and employs red-teaming and bug bounties.

In practice

• Implement tiered access for powerful AI models.
• Use red-teaming to test AI safety bypasses.
• Route sensitive queries to restricted models.

Topics

• Anthropic Fable 5
• AI Model Safety
• Cybersecurity Vulnerabilities
• Project Glasswing
• AI Model Pricing
• Red Teaming

Best for: AI Engineer, Machine Learning Engineer, CTO, Tech Journalist, AI Security Engineer, Director of AI/ML

Related on AIssential

• Anthropic offers EU access to Mythos — Anthropic's Mythos AI highlights dual-use risks, prompting calls for AI development slowdowns and international safety regulations.
• Anthropic's new Claude Fable 5 is the same base model as Mythos but with guardrails attached — Fable 5 offers Mythos-class AI power with integrated safety guardrails for broader, secure general use.
• Anthropic Claude Fable 5 — Anthropic's new Claude Fable 5 offers superior agentic capabilities but introduces controversial, silent performance limitations for specific research tasks.
• Why Anthropic believes its latest model is too dangerous to release — Advanced LLMs like Mythos Preview demonstrate unprecedented autonomous cyberoffense capabilities, fundamentally altering the cybersecurity landscape.
• BREAKING: Anthropic’s “Dangerous” AI Model Mythos LEAKED — Controlling access to powerful AI models is as critical as the models' inherent safety.
• Why the world’s banks are so worried about Anthropic’s latest AI model — Anthropic's Mythos AI model reveals critical cybersecurity vulnerabilities, heightening bank concerns over AI-driven cybercrime.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI (artificial intelligence) | The Guardian.