Anthropic Releases Public Version of Fable 5 With Automated Safety Guardrails
Key Takeaways
- ▸Fable 5 is the first Mythos model released to the general public, with automatic routing of sensitive queries to Opus 4.8 to prevent misuse in cybersecurity and chemical/biological domains
- ▸Anthropic has conducted over 1,000 hours of red-teaming and ran a bug bounty program with no successful bypass methods discovered, demonstrating substantial safety engineering
- ▸Unrestricted access to Mythos 5 remains limited to ~200 select organizations; the company identified large-scale attempts to extract the technology for training competing models in authoritarian countries
Summary
Anthropic has released Fable 5, the first model from its advanced Mythos class, to the general public with carefully implemented safety restrictions. The model was announced in April but restricted to partner institutions for months due to cybersecurity concerns about its ability to identify vulnerabilities in critical infrastructure like banking platforms and power grids. Queries about sensitive topics such as cybersecurity, biology, and chemistry are automatically routed to the less powerful Opus 4.8 model. In parallel, an unrestricted version, Claude Mythos 5, remains available only to approximately 200 select organizations in over 15 countries enrolled in Anthropic's Project Glasswing program, which is expected to expand further.
The company has implemented substantial safeguards including red-teaming by outside experts who spent over 1,000 hours attempting to bypass restrictions, and a bug bounty program that found no complete unlock method. Anthropic is deploying Mythos 5 in collaboration with the U.S. government as part of a White House arrangement to test leading AI companies' most powerful models before release. The release comes amid significant geopolitical tension, as the Trump administration has pressured Anthropic to lift restrictions on mass surveillance and autonomous lethal weapons—leading the Pentagon to sever its defense contracts. Fable 5 is priced at a premium: $10 per million input tokens and $50 per million output tokens, double the cost of Opus 4.8, reflecting Anthropic's substantial computing costs, including a $1.25 billion monthly datacenter lease.
- Pentagon severed Anthropic defense contracts over the company's refusal to support mass surveillance and autonomous lethal weapons, highlighting geopolitical constraints on AI safety positions
- Fable 5 pricing ($10/$50 per million tokens) is double Opus 4.8 and reflects premium positioning as Anthropic pays substantial computing costs, including $1.25bn monthly for xAI datacenter access
Editorial Opinion
Anthropic's release of Fable 5 demonstrates a nuanced approach to capability deployment—restricting the most powerful model while making it available to the broader developer community, a pragmatic middle ground between hoarding and recklessness. The company's deep collaboration with U.S. government testing and transparent red-teaming signal genuine commitment to responsible scaling, though the effectiveness of automated query routing against sophisticated adversaries remains an open question. The geopolitical pressure from the Trump administration reveals a uncomfortable truth: corporate safety commitments can clash with state interests, and even well-intentioned AI companies lack autonomy when national security concerns take precedence. Anthropic's premium pricing suggests confidence in differentiation, but in a market where Claude faces intensifying competition from GPT-4o, Llama, and others, whether developers will pay double for safety guardrails remains untested.
