By EGC Group For the first time in nearly seven years, a leading AI company has decided its new model is too capable of giving to the public. Anthropic confirmed this month that it has been training and testing a new model called “Claude Mythos,” describing it as “a step change” and “the most capable we have built to date.” The model’s existence was first revealed at the end of March, when security researchers discovered Anthropic had inadvertently left a draft blog post and thousands of unpublished documents in a publicly searchable data store. Before the material was pulled down, over 3,000 internal documents were exposed through the CMS misconfiguration. Presenting “Project Glasswing” Instead of a public release, Anthropic is giving access to over 50 tech organizations, including Microsoft, Nvidia, and Cisco, via an initiative called “Project Glasswing,” with over $100 million in usage credits. Why the restricted release? Claude Mythos Preview demonstrates a leap in cybersecurity capabilities. The vulnerabilities it has spotted have in some cases survived decades of human review and millions of automated security tests. Frontier AI models are now becoming competitive with the best human experts at finding and exploiting software vulnerabilities. This model is powerful enough to identify undiscovered flaws and, in some cases, build working exploits for them. As one example, Anthropic found its cybersecurity capabilities were surprisingly advanced in contrast to previous models. This discovery led directly to the creation of Project Glasswing. Then vs. Now The last time something like this happened was 2019, when OpenAI withheld GPT-2 over concerns about misuse. That model now looks unremarkable. The gap between “too dangerous” in 2019 and “too dangerous” in 2026 is a useful way to appreciate how fast the capability curve is genuinely progressing. For marketing and business leaders, the Mythos story is not really about cybersecurity. It is a calibration moment. Anthropic chose the name from the Greek word mythos, a foundational narrative that shapes understanding of reality. That framing is deliberate. The AI tools that your teams are already using for content, media, strategy, and GEO visibility sit on the same underlying architecture as a model that just identified zero-day vulnerabilities in every major operating system and browser on earth. Experimentation is Over The organizations treating AI as a pilot-phase experiment in 2026 are making a bet that the capability curve will slow down long enough for them to catch up. At CES earlier this year, the consensus among agency and brand leaders was clear: The time for experimentation has ended. AI is no longer a test. It is becoming part of the industry’s infrastructure, and both brands and agencies are being measured on how effectively they deploy it. Mythos is a data point on where that infrastructure is heading. The question for leaders is not whether this level of AI capability will eventually touch their businesses. It is whether these leaders are either building toward it or waiting to react when it arrives.
