On June 9, Anthropic announced the rollout of its public version of the Mythos AI model, now known as Claude Fable 5, which includes guardrails to prevent its use in sensitive areas like cybersecurity. This decision follows a preview that highlighted Mythos’s ability to uncover thousands of software vulnerabilities, causing significant concern globally. To ensure safety, extensive testing was conducted to prevent users from manipulating the model for inappropriate tasks, such as identifying cyber vulnerabilities. As the company aims to build on its momentum and compete with rival OpenAI, it plans to gradually expand access through a systematic trusted-access program. The pricing for the new model is set at $10 per million input tokens and $50 per million output tokens.

Mythos: Mythos is Anthropic’s advanced AI model first previewed earlier in the year, noted for its ability to identify software vulnerabilities. The public release includes new restrictions that block requests related to cybersecurity while allowing other capabilities. Preview users can upgrade to the guarded version of the model.
Anthropic: Anthropic is an AI research company developing large language models in the Claude family with a focus on safety and capability. It is rolling out a public version of its Mythos model under strict guardrails that prevent use in cybersecurity or other high-risk domains. The startup is expanding model access beyond limited partners like the U.S. government to support broader commercial use.
Dianne Penn: Dianne Penn is Anthropic’s head of product management, research and labs. She described how the new model handles restricted requests by refusing them and defaulting to safer alternatives. Her statements outline the testing process used to prevent guideline bypasses.
Claude Fable 5: Claude Fable 5 is Anthropic’s most powerful model released for wider public availability, emphasizing performance in software engineering and analytics. It incorporates safety mechanisms that refuse risky queries and fall back to prior model versions when needed. The model is positioned as more efficient for complex tasks compared with earlier releases.

`json
{
“Model Release”: “Anthropic is introducing its latest model for broader use while ensuring it is not usable in cybersecurity.”,
“Safety Measures”: “Extensive testing ensures the model cannot be manipulated to perform actions outside its allowed guidelines.”,
“Access Expansion”: “The company plans to expand availability through a systematic trusted-access program over time.”
}
`