Anthropic and OpenAI Develop Selective Access Strategies for AI Models

<p>AI labs are adopting a strategy of selective access to control their advanced models while commercializing them. This approach is exemplified by OpenAI's trusted-access program and a forthcoming program from Anthropic, which aim to establish a new framework in cybersecurity by determining which entities can utilize advanced AI capabilities.</p><p><strong>Details:</strong> Anthropic announced on June 9, 2026, that it will make a version of its Mythos class of models, known as Fable 5, available to the public. Fable 5 includes safeguards that prevent certain high-risk requests related to cybersecurity and biology, directing those inquiries to another model, Claude Opus 4.8.</p><ul><li>Anthropic is also providing an upgrade to its Mythos 5 model for users of its restricted Mythos Preview program.</li><li>Dianne Penn, Anthropic's head of product management for research and labs, indicated that the company is taking a cautious approach at launch, which may result in some legitimate security inquiries being redirected away from Fable 5.</li></ul><p>Anthropic is developing a formal trusted-access program to determine who will have access to Mythos 5 and future models with fewer restrictions. The timeline for this program has not been disclosed. In recent months, various organizations have been lobbying for access to the Mythos Preview program, and last week, Anthropic expanded access to over 150 companies and government entities.</p><p><strong>Context:</strong> OpenAI has implemented a similar two-tier access system, vetting security researchers and organizations to grant access to models that can enhance their cybersecurity efforts. The company has released an alternate version of its GPT-5.5 model with fewer restrictions to assist cyber defenders in identifying vulnerabilities and analyzing malware.</p><p>As AI labs navigate selective access, they are tasked with determining who can utilize the most advanced cybersecurity capabilities. Security vendors, researchers, and critical infrastructure operators are actively seeking access to these frontier AI technologies.</p><p><strong>Implications:</strong> This selective access model allows Anthropic and OpenAI to manage the deployment of powerful AI capabilities while exploring monetization strategies as they consider entering public markets.</p><p><strong>Future Considerations:</strong> It remains to be seen whether users granted trusted access will uncover vulnerabilities and develop products that outperform those available to organizations without access.</p>