Anthropic: Glasswing grows, Mythos soon for all Apfelpatient

Anthropic is opening its previously tightly controlled security program, Project Glasswing, to significantly more partners and simultaneously promising to offer its most powerful Mythos models to all customers soon. This move coincides remarkably closely with the AI company's plans for an IPO.

When Anthropic unveiled its AI model Claude Mythos in April, access was initially restricted to a small circle of select partners – including Apple, which deployed the technology as part of a cybersecurity initiative. This controlled release is part of Project Glasswing, a program through which Anthropic aims to channel the model's particularly sensitive capabilities in a more responsible manner. Shortly after its launch, Mythos was linked to the discovery of security vulnerabilities in macOS – an indication of the technology's effectiveness in vulnerability research. This very program is now experiencing significant growth, and Anthropic is simultaneously announcing broader market access.

Approximately 200 organizations are participating in the security program

Anthropic announced that, following several weeks of close collaboration with existing partners, the security industry, open-source software maintainers, and the US government, it is expanding Project Glasswing to include approximately 150 new organizations. Each of these organizations must first meet the company's security requirements before gaining access. With this planned expansion, the partner list is expected to grow to around 200 organizations.

The program now includes partners in more than 15 countries, with a stronger presence of companies from the communications and hardware sectors. According to a Financial Times report, Samsung and NATO are among the new participants. The requirement for each partner to meet predefined security criteria has a solid basis: There have already been reports of unauthorized access to Mythos, underscoring the sensitive nature of the model.

Mythos class should be open to all customers

Perhaps the most significant point is the announcement that Mythos-class models will be generally available in the foreseeable future. Just last week, during the unveiling of the Claude Opus 4.8 model, Anthropic had already stated its intention to release these particularly high-performance models to all customers soon.

However, stronger cybersecurity measures are needed before general release, as models in this performance class carry a higher risk potential. Anthropic states that it is making rapid progress in developing these protection mechanisms and expects to be able to roll out the Mythos class widely within the coming weeks. This would transform an exclusive partner technology into a generally available product within a few months.

Closely linked in time to the IPO

The expansion of Glasswing comes just one day after Anthropic publicly announced its plans to go public later this year. This timing is remarkable: A company preparing for an IPO demonstrates both its reach and a responsible approach to high-risk technology by opening up a previously exclusive program. The competition has already reacted to Glasswing – a rival offering recently emerged explicitly as a counterpoint to Anthropic's program.

From closed testing to widespread availability

Within just a few weeks, Anthropic shifted its Mythos strategy from a carefully curated partner program to a model intended to be open to everyone. The parallel expansion of Glasswing demonstrates that the company is not losing sight of security concerns, but rather linking broader access to clear conditions. Whether the announced safeguards will prove sufficient in practice will only become apparent once the first regular customers gain access to the Mythos class. (Image: Shutterstock / Primakov)