News

Anthropic Expands Project Glasswing Access to Mythos While Broadening Reach of Fable 5

• By John K. Waters
• 06/11/2026

As governments and businesses grapple with the risks posed by increasingly capable AI systems, Anthropic is expanding access to one of its most powerful models, but only under tightly controlled conditions.

The company announced it is broadening participation in Project Glasswing, a program that provides vetted organizations with access to Claude Mythos, Anthropic's restricted, cybersecurity-focused frontier model. The expansion adds roughly 150 organizations across more than 15 countries, including entities involved in critical infrastructure sectors such as energy, water, healthcare, communications, and hardware.

The move reflects a growing tension within the AI industry. Developers of advanced models increasingly face pressure to make their technologies useful to organizations responsible for defending critical systems, while also limiting access to capabilities that could be misused by malicious actors.

Anthropic launched Project Glasswing as a controlled-access initiative intended to give selected organizations the opportunity to evaluate and use advanced cybersecurity capabilities without making the underlying model broadly available.

In announcing the expansion, the company said the program is intended to help organizations "strengthen the resilience of critical infrastructure" and improve their ability to identify and respond to emerging cyber threats.

The expansion does not mean that Claude Mythos is becoming generally available. Anthropic continues to restrict access to the model, which is designed for advanced cybersecurity applications and has undergone extensive safety evaluations.

Instead, the company is pursuing a two-track strategy. While Mythos remains limited to approved participants, Anthropic recently introduced Claude Fable 5, which the company describes as a Mythos-class model with additional safeguards and restrictions.

According to Anthropic, Fable 5 is derived from the same technology lineage as Mythos but incorporates controls intended to limit certain capabilities related to cybersecurity, biology, and chemistry. The goal is to make more advanced AI capabilities available to a broader audience while maintaining safeguards around potentially sensitive applications.

The distinction highlights a challenge facing AI developers. As models become more capable, companies must decide which capabilities can be safely released and which should remain subject to additional oversight.

Anthropic has positioned itself as one of the industry's most vocal advocates of staged deployment and model safety. The company has argued that increasingly powerful systems require corresponding advances in governance, evaluation, and access controls.

Project Glasswing offers a practical example of that philosophy. Rather than treating access as an all-or-nothing decision, Anthropic is using a tiered approach that grants broader availability to safeguarded models while reserving frontier cybersecurity capabilities for vetted organizations.

The initiative also arrives as AI companies face growing scrutiny from regulators, policymakers, and security researchers regarding how advanced models are deployed and monitored. Questions surrounding accountability, access controls, and dual-use technologies have become increasingly central to discussions about the future of artificial intelligence.

For Anthropic, the expansion of Project Glasswing suggests that the company sees a role for advanced AI systems in strengthening cybersecurity defenses, provided those systems are deployed within carefully managed governance frameworks.