Anthropic announced Project Glasswing last week with the kind of language reserved for genuine inflection points. Claude Mythos Preview, a frontier model the company deliberately chose not to release publicly, had already identified thousands of zero-day vulnerabilities across every major operating system and web browser — including a bug that had been sitting undetected in OpenBSD for 27 years. The partner list reads like a who’s who of critical infrastructure: Microsoft, Nvidia, AWS, Apple, Google, CrowdStrike, Palo Alto Networks, Cisco, Broadcom, JPMorgan Chase, and the Linux Foundation. Anthropic committed $100 million in usage credits and framed the whole initiative as an urgent attempt to give defenders a head start before these capabilities proliferate beyond actors committed to deploying them safely.

It is an impressive announcement. It is also solving the wrong problem.

Defenders have never lacked for vulnerabilities to find. The global security industry produces an extraordinary volume of vulnerability intelligence every single day — from CVSS-scored CVEs to proprietary threat feeds to penetration testing engagements to bug bounty programs to automated scanning tools that have been running continuously for years. The problem has never been the discovery pipeline. The problem has always been what happens after discovery, and that is where Project Glasswing goes conspicuously quiet.

Consider what the average enterprise security team faces. At any given moment, a mid-sized organization with moderately complex infrastructure carries thousands of known, unpatched vulnerabilities in production systems. Security teams know about them. They have known about many of them for months, sometimes years. The vulnerabilities sit in queues not because defenders are lazy or incompetent, but because remediation requires coordination across organizational boundaries that security teams do not control — development pipelines, change management windows, vendor patch cycles, legacy systems that cannot be patched without breaking something else, and business owners who will not accept downtime for a vulnerability rated high rather than critical. Prioritization is not a science. It is a negotiation, and security teams lose that negotiation constantly because the business pressure to keep systems running always outweighs the abstract risk of a vulnerability that attackers have not yet exploited.

Now Anthropic is proposing to dramatically accelerate the front end of a pipeline that is already choking at the back end. Mythos Preview will find more vulnerabilities, faster, with greater sophistication than any tool that preceded it. The model can chain three, four, or five vulnerabilities in sequence to construct exploits that would have required elite human expertise to discover manually. That capability is genuinely remarkable. It is also about to make the remediation problem catastrophically worse for every security team that does not have the operational capacity to act on what the model finds — which is most of them.

This is not a hypothetical concern. Alex Stamos, who has spent his career at the intersection of security engineering and institutional reality, put a sharp timestamp on the broader capability race: roughly six months before open-weight models catch up to frontier models in vulnerability discovery, at which point the exploit pipeline becomes democratized and every malicious actor on the planet gains access to the same discovery capability Anthropic is currently restricting to forty-plus vetted organizations. That six-month window is Anthropic’s argument for urgency. It is also an argument for asking a harder question: if defenders cannot remediate what they already know about, what exactly does finding more vulnerabilities faster accomplish for them?

The honest answer is that it accomplishes a great deal for the forty-plus organizations in the Glasswing coalition, most of which have the engineering headcount, the operational maturity, and the organizational authority to act on what the model surfaces. Microsoft can patch Windows. Google can fix Chrome. Palo Alto Networks can harden its own platform. For those organizations, Mythos Preview delivers genuine defensive leverage. The rest of the ecosystem — the thousands of enterprises, municipalities, healthcare systems, and critical infrastructure operators running on the software these companies produce — will inherit better-patched upstream dependencies without receiving any of the capability that produced them.

What the security industry needs alongside a more powerful vulnerability discovery engine is a more powerful remediation operating model — one that uses the same agentic AI capabilities to triage, prioritize, and orchestrate fixes at machine speed rather than at the speed of a change management committee. The bottleneck in enterprise security has never been intelligence. It has been action. Remediation requires touching production systems, coordinating across organizational silos, and accepting risk in the short term to reduce risk over time. Those are fundamentally human organizational problems, and no amount of vulnerability discovery acceleration changes that calculus without a corresponding investment in remediation capacity.

Anthropic deserves credit for recognizing that these capabilities require governance before they require democratization, and Project Glasswing is a serious attempt to structure that governance around the organizations best positioned to absorb it. But treating discovery as the primary leverage point reflects a model of the defender’s problem that has not been accurate for at least a decade. The vulnerability queue is not empty. It is overflowing. Adding more to it faster, even with the best intentions, is not a security strategy — it is a faster way to drown.