How Was CrowdStrike Even Invited to Glasswing?
The Glasswing announcement was supposed to be a moment of industry leadership. What it actually revealed is that the institutions pay plans cannot tell you whether cybersecurity outcomes are something their executives are accountable for.
Jin S. Chun, CFA, FRM — Managing Partner, CAT Ventures LLC | April 2026
On April 7, Anthropic released a preview of its new Claude Mythos model, accompanied by Project Glasswing — an initiative naming twelve launch partners who will work with Anthropic on the cybersecurity implications of AI capabilities that, by Anthropic's own admission, the model was not specifically trained for. Among the twelve partners is CrowdStrike. The same CrowdStrike whose July 19, 2024 software update grounded airlines, paralyzed hospitals, and produced an estimated 5.4 billion dollars in direct financial losses to Fortune 500 companies1 — the most consequential single-vendor cybersecurity failure in history. The question is not rhetorical. How did the firm responsible for that outage end up at the table where the next generation of cybersecurity governance is being defined?
The answer is in CrowdStrike's most recent proxy statement, and it is more interesting than the question.
CrowdStrike holds its shareholder vote on executive pay only once every three years. The most recent vote, in June 2024, came in at 86.5 percent approval — below the 90 percent threshold that typically prompts the compensation committee to reach out to unhappy shareholders. That vote took place one month before the July 19 outage. Because of the triennial cycle, the next vote will not occur until approximately June 2027 — three full years after the incident, with no shareholder mechanism to register dissatisfaction in the interim. The compensation plan in effect on July 19 gave 80 percent weight to new subscription growth and 20 percent weight to operating profit. None of it measured whether the software was safe to deploy.2
This is not a CrowdStrike-specific observation. I recently pulled the executive pay disclosures for ten pure-play cybersecurity vendors and nine property and casualty insurers writing material commercial cyber coverage. Nineteen companies whose entire business is cybersecurity in one form or another. Across all nineteen, the number that link executive pay to a measurable cybersecurity outcome is zero. The compensation metrics are, without exception, financial: revenue, subscriptions, operating profit, combined ratio. An industry's pay plan is its most legally precise public statement about which outcomes it holds its executives accountable for. By that standard, the cybersecurity industry is telling us that cybersecurity outcomes are not on the list.
This is predictable, not accidental. Two decades ago, economists Ross Anderson and Tyler Moore identified the structural reason: the executive making the security investment decision is rarely the party who pays when that decision proves wrong. The cost of a breach lands on customers, on counterparties, on the broader financial system. None of that cost shows up in the executive's pay plan. Anderson and Moore put it plainly: 'vendors have succeeded in dumping most software risks, but this outcome is also far from being socially optimal.' The zero-out-of-nineteen finding is the twenty-year downstream confirmation of their prediction.3
The comparison to other industries is instructive. Every major oil and gas producer ties a meaningful portion of executive pay to process safety events. Every US Class I railroad ties weight to derailment rates and reportable incidents. Major mining companies use injury rates as board-level compensation inputs. Every major international shipping company operates under standards that feed directly into executive pay scorecards. At Paul O'Neill's Alcoa, tying executive pay to worker safety caused the injury rate to fall from 1.86 in 1987 to 0.2 by 2000 while market capitalization rose from three billion dollars to twenty-seven billion. After the Upper Big Branch mine disaster in 2010, the CEO of Massey Energy was convicted in federal court of conspiring to violate mine safety standards and served a year in federal prison.4 None of these industries has solved safety. All of them have decided that the people running the company should be paid based on whether the risk they exist to manage is actually being managed. The cybersecurity industry has not made that decision. Neither has the commercial cyber insurance industry.
The deeper problem is that compliance documentation cannot see what matters. The risk lives in the gaps. Those who lived through the 2008 financial crisis will recognize what the companion working paper calls counterparty blindness — the condition in which each participant has complete visibility into their own position and zero visibility into the correlated exposures of others. The biological term for the connective tissue between organs that no single organ's diagnostic can detect is interstitial. In financial structures, we simply call it the gap. The gap is where the risk lives, and it is exactly where compliance documentation is not designed to look.5
Six months ago, First Brands Group filed for Chapter 11 bankruptcy. Creditors are now alleging that as much as 2.3 billion dollars in short-term financing has, in the words of one court filing, "simply vanished."6 The apparent mechanism: the same trade receivables pledged to multiple lenders who did not know each other existed. Jefferies' Point Bonita Capital Fund had 715 million dollars in receivables payable by Walmart, AutoZone, NAPA, O'Reilly, and Advance Auto Parts — receivables that had paid on time and in full for almost six years before the music stopped on September 15. UBS O'Connor had over 500 million in exposure. Raistone, Millennium, a dozen CLOs, and multiple other lenders all had positions. Each did its credit work. Each had complete documentation. None was structured to compare notes with the others. The only entity with the complete picture was First Brands itself, sitting at the center as the servicer, directing the payments. When the First Brands lawyer was asked at the October 1 hearing where the cash had gone, he answered: "It's not here."
Six months ago, First Brands Group filed for Chapter 11 bankruptcy. Creditors are now alleging that as much as 2.3 billion dollars in short-term financing has, in the words of one court filing, "simply vanished."5 The apparent mechanism: the same trade receivables pledged to multiple lenders who did not know each other existed. Jefferies' Point Bonita Capital Fund had 715 million dollars in receivables payable by Walmart, AutoZone, NAPA, O'Reilly, and Advance Auto Parts — receivables that had paid on time and in full for almost six years before the music stopped on September 15. UBS O'Connor had over 500 million in exposure. Raistone, Millennium, a dozen CLOs, and multiple other lenders all had positions. Each did its credit work. Each had complete documentation. None was structured to compare notes with the others. The only entity with the complete picture was First Brands itself, sitting at the center as the servicer, directing the payments. When the First Brands lawyer was asked at the October 1 hearing where the cash had gone, he answered: "It's not here."
That is the same structural pattern as the cybersecurity compensation problem. The compliance documentation is complete. Every box is checked. Every certification is current. The risk is not in any of the boxes. It is in the gaps — the spaces between the rooms that compliance was never designed to connect. The CrowdStrike compensation committee did not fail to do its work. It did its work thoroughly, and its work could not see the thing that mattered, because the thing that mattered was not in any field on any form the committee was designed to look at.
Anthropic had the leverage to make this different. It resisted Department of Defense applications on safety grounds. It publishes more detailed system cards than any competitor. And yet it assembled a coalition for its most dangerous model release without apparently asking whether its partners' compensation architectures hold executives accountable for the outcomes Mythos makes possible. The lab's safety framework is technically sophisticated and institutionally naive. For a lab serious enough about safety to push back on the DoD, that gap is worth closing — and it is a gap Anthropic is uniquely positioned to close. Making governance accountability a condition of Glasswing participation would cost nothing and change everything about what kind of initiative Glasswing actually is.
The Glasswing announcement was supposed to be a moment of industry leadership. What it actually revealed is that the institutions defining the response to the most significant capability shift in cybersecurity history are the same institutions whose pay plans cannot tell you whether cybersecurity outcomes are something their executives are accountable for. That is not a failure of any individual compliance function. It is a structural failure of an industry approaching a threshold that every other essential service industry has eventually crossed. When the market failure becomes too large and too systemic for voluntary governance to contain, the historical answer is not more coalitions. It is utility regulation. The cybersecurity industry may not be ready to acknowledge that yet. July 19, 2024 was the first bill. It will not be the last.
About this piece
This piece is adapted from the author's working paper The Gaps: Governance Physics and the Cybersecurity Capability Break (CAT Ventures LLC, April 2026).
Jin S. Chun, CFA, FRM, is Managing Partner of CAT Ventures LLC: jin@catventures.com.
Notes
1. CrowdStrike DEF 14A filed May 6, 2024; 8-K filed June 21, 2024 (EDGAR CIK 0001535527). Triennial cycle and STI KPI weights from the Compensation Discussion & Analysis section.
2. Anderson, R. and Moore, T. (2006). The Economics of Information Security. Science, 314(5799), 610–613. Free PDF: tylermoore.utulsa.edu/science-econ.pdf
3. Full treatment of counterparty blindness, the First Brands structure, the regulatory genealogy of proxy advisory services, and the SGF framework is in the companion working paper cited above.
4. The $5.4 billion direct financial loss estimate is from Parametrix, published July 24, 2024. The figure covers Fortune 500 companies excluding Microsoft. Insured losses were estimated at $540 million to $1.08 billion.
5. First Brands Group LLC, Chapter 11 Case No. 25-11606 (Bankr. D. Del.). The $2.3 billion figure and "simply vanished" language from Raistone emergency motion. "It's not here" from court transcript, October 1, 2025 hearing. Point Bonita $715M, UBS O'Connor $500M+, and Raistone $631M figures from creditor disclosures and court filings.
6. Don Blankenship, CEO of Massey Energy, was convicted in December 2015 of conspiring to willfully violate federal mine safety standards following the April 5, 2010 Upper Big Branch mine disaster that killed 29 miners. He served one year in federal prison. DOJ public record.