During a session at the AI Impact Summit in New Delhi on February 19, 2026, Anthropic CEO Dario Amodei participated in a meeting with French President Emmanuel Macron. The recent exposure of Anthropic’s secretive AI model named Claude Mythos occurred after confidential details surfaced in a publicly accessible database. Consequently, the company has revealed plans to collaborate with major global enterprises to utilize this model in identifying potential security weaknesses within their systems.
The initiative, known as Project Glasswing, involves a controlled deployment of Mythos to approximately 40 organizations for preview purposes. This version of the model is claimed to excel at pinpointing software vulnerabilities, surpassing most human experts in this field. Notable partners supporting this project include Amazon Web Services, Apple, Google, JPMorganChase, Microsoft, and NVIDIA.
Anthropic reports significant early success from this partnership, having identified numerous critical vulnerabilities across various operating systems and web browsers. The company envisions Mythos as a transformative force in the realm of cybersecurity based on its promising performance compared to Claude Opus 4.6 in benchmark tests.
Anecdotal evidence further supports the effectiveness of Mythos, such as discovering a long-standing bug in OpenBSD and uncovering a series of vulnerabilities in Linux capable of compromising entire systems. Despite initial concerns about the model’s potential risks to cybersecurity, Anthropic has chosen not to release Mythos Preview to the public.
This shift from extreme caution due to perceived dangers to active deployment within critical technology infrastructure marks a notable transition for Anthropic and its approach toward Mythos. The company’s strategic positioning of the model appears influenced by past instances of exaggerated claims within the AI industry.
Reflecting on previous instances like Elon Musk’s OpenAI text-generation tool development, which was initially deemed too risky for release but eventually made public without catastrophic consequences, underscores the evolving nature of AI technologies and their impact on various sectors. Anthropic’s evolution from promoting Claude Opus 4.6’s capabilities to unveiling Mythos reinforces the dynamic landscape surrounding AI advancements and their role in shaping cybersecurity practices.
Considering the ongoing cycle of AI innovation and adaptation within cybersecurity frameworks, models like Mythos are poised to significantly influence future security measures despite inherent challenges associated with their continuous refinement and operation.