Five Reasons Anthropic Kept Its Cybersecurity Breakthrough Invite-Only

Anthropic launched Project Glasswing, a cybersecurity initiative built around the unreleased frontier model Claude Mythos Preview, which the company says has identified thousands of significant vulnerabilities across major operating systems and web browsers. Backed by $100 million in model usage credits and a coalition that includes Amazon, Apple, Google, Microsoft, Nvidia, CrowdStrike, Palo Alto Networks and the Linux Foundation, this is a consequential cybersecurity breakthrough from an AI company and a departure from how proprietary frontier models have been released so far.

This announcement cannot be understood in isolation. It arrived on the same week Anthropic disclosed a $30 billion annualized revenue run rate and sealed a multi-gigawatt compute deal with Google and Broadcom. There are at least five explanations for the restricted rollout.

Five Reasons Behind The Cybersecurity Breakthrough's Rollout

A real jump in capability. The evidence is difficult to dismiss. Mythos Preview chains together vulnerabilities that individually appear benign but collectively yield complete system compromise, linking three, four or sometimes five flaws into exploit sequences no prior automated tool had produced. Nicholas Carlini, a recognized expert in adversarial machine learning who has been working with the model, coauthored a technical evaluation highlighting that "over 99% of the vulnerabilities we've found have not yet been patched." In one case, the model found a 27-year-old bug in OpenBSD, an operating system built with security as its founding principle.

Responsible AI governance shaping deployment. Anthropic has spent the past several months in a dispute with the Pentagon over military uses of Claude. The conflict escalated when the government designated Anthropic a supply-chain risk, a step later blocked by a federal judge. Project Glasswing follows the same basic logic: if you build something powerful enough to be dangerous, you control how it enters the world.

Strategic marketing through scarcity. A model described as too dangerous to release publicly generates enormous press attention. That narrative was reinforced by reports last month that revealed Mythos through a leak and intensified attention around the model. The timing, framing and partner selection are calibrated to position Anthropic at the center of the national security conversation at a moment when the company is considering an IPO later this year.

A phased rollout driven by capacity constraints. Anthropic is still constrained by computing capacity. The company has throttled usage limits during peak hours, and its enforcement actions against third-party tools accessing Claude Code drew sharp criticism from developers, including Ruby on Rails creator David Heinemeier Hansson, who called the moves "very customer hostile." When demand exceeds infrastructure, a restricted rollout to a group of organizations is operationally necessary. The 3.5 gigawatts of new compute capacity announced this week will not arrive until 2027.

Premium pricing that reveals the business model. This is the most telling indicator. After the $100 million credit period, Claude Mythos Preview will cost $25 per million input tokens and $125 per million output tokens. That is five times the price of Anthropic's own Claude Opus 4.6 at $5 and $25. Cross-company comparisons are harder because pricing varies by context window, service tier and caching rules, but Mythos is clearly being positioned as a premium, access-restricted security product. This spread reflects demand from customers willing to pay a significant premium for vulnerability detection capabilities that no alternative can match.

What The Industry Is Saying About This Cybersecurity Breakthrough

Jim Zemlin, CEO of the Linux Foundation, noted that open-source software makes up most of the code in modern systems, yet its maintainers have historically lacked access to sophisticated security tools. Project Glasswing, he said, "offers a credible path to changing that equation."

Simon Willison, a widely followed programmer and technology commentator, wrote that the restricted deployment approach "sounds necessary," while noting it would be beneficial for OpenAI to participate as well. If the capabilities Anthropic describes proliferate across the industry, restricting Mythos to a coalition of partners is a temporary measure at best.

Reporting by Michael Nuñez, the Editorial Director of VentureBeat, framed the announcement in financial terms, observing that a high-profile, government-adjacent cybersecurity initiative with blue-chip partners strengthens the company's IPO narrative. The scale of computing required to run a model like Mythos Preview also helps explain why Anthropic is pairing the launch with enterprise partnerships and new infrastructure deals.

The Cybersecurity Breakthrough In Context: What Comes Next

Anthropic's latest numbers suggest it is ahead of OpenAI in enterprise deployment and is collecting the benefits of this strategy. The $30 billion run rate, up from $9 billion at the end of 2025, shows that the company has surpassed OpenAI in annualized revenue. More than 1,000 business customers now spend over $1 million annually, a figure that doubled in under two months. Anthropic's revenue is overwhelmingly enterprise-driven, which carries higher retention and better expansion economics than OpenAI's consumer-heavy composition.

The company faces real scrutiny on two fronts. Pricing friction is recurring; subscription-tier users have pushed back on throttled limits and opaque quota mechanics. Capacity constraints compound the issue. CEO Dario Amodei, in an interview with Dwarkesh Patel, warned that even a slight miscalculation in compute investment timing could sink the company, saying there is "no hedge on earth" that could prevent bankruptcy if revenue projections fall short of commitments. That discipline has served Anthropic financially, but it creates operational pain for developers and enterprise teams who need predictable, continuous access.

Project Glasswing is a bet that the controlled rollout of a cybersecurity breakthrough will produce value for the company while containing the risk of the new capabilities reaching adversary nations or non-state actors. It is a bet that the market will reward Anthropic for the restraint. Initial reactions suggest Anthropic has, at minimum, succeeded in framing the debate on its own terms. Whether the gradual, private rollout strategy delivers on its full promise will depend on execution, transparency and the progress made by other frontier labs.