Anthropic has officially launched two new AI models, Claude Mythos 5 and Claude Fable 5, with distinct audiences in mind. While Mythos 5 is available exclusively to select industry partners, Fable 5 is set to be publicly accessible. The company has made it clear that these new versions possess advanced capabilities compared to the previous Mythos Preview model, which was initially shared under a limited release due to concerns over potential misuse by malicious actors.
The release of Claude Mythos 5 comes with targeted distribution aimed primarily at partners within the Project Glasswing consortium, which is a collaboration involving the US government. In contrast, Claude Fable 5 will implement various restrictions that prevent it from handling specific inquiries related to cybersecurity and sensitive scientific topics, directing such queries instead to an older AI model, Claude Opus 4.8. This is a precautionary measure to mitigate risks associated with the sophisticated functionalities of the new AI models.
Diane Penn, Anthropic’s head of product management, acknowledged that the company has been actively considering how to manage the vulnerabilities that come with the advanced capabilities of Mythos. Testing and user feedback obtained since the initial launch in April have refined their approach. Currently, the model’s protective mechanisms prioritize caution, meaning some benign queries may still be redirected to the less advanced model as a safety net.
In addition to licensing the Claude Mythos 5 exclusively to certain industry partners and select biology researchers, the company hinted at plans to enlarge access through a trusted access program in the future. The need for robust safeguards against misuse remains crucial, as these AI models can potentially design hacking tools capable of exploiting software vulnerabilities.
Anthropic has priced Claude Fable 5 and Mythos 5 at $10 per million input tokens and $50 per million output tokens, which is higher than their previous publicly available models but cheaper than the Mythos Preview. The cautious rollout speaks to the underlying concerns over cybersecurity that have plagued the AI industry, particularly as both Anthropic and competitors like OpenAI race toward public initial offerings while navigating these technical and ethical challenges.
The ongoing effort to secure the software landscape against AI-driven threats is underscored by Anthropic’s commitment to ensuring their models are developed with safeguards that can prevent exploitation, paving the way for a future where such technologies are adopted more broadly. More information about the project and updates can be found on Anthropic’s official blog and the Project Glasswing initiative.
