Yesterday's [last week's] Microsoft outages and a wave of voice-phishing campaigns targeting major financial firms arrived as a pointed reminder. Hudson Attar and Arham Habib, writing for Bridgewater Associates1, are unambiguous about what these episodes represent: "one flavor of disruption that is likely to become more common as the cyber-exploit exponential we write about in this piece continues to unfold." They are not waiting for confirmation of malicious intent before drawing portfolio conclusions. The firm says it is "managing money based on a one-in-three chance of an economically destabilizing attack within the next year."
That is a headline number worth sitting with. One-in-three is not a tail risk. It is a scenario that belongs inside any serious risk framework for advisors and institutional investors alike.
An Arms Race With Asymmetric Stakes
The analytical core of the report rests on a structural asymmetry. Defenders must patch every vulnerability in a code base that stretches back decades. Attackers need find only one. That imbalance has been dramatically sharpened by AI. Attar and Habib write: "AI-enabled cyber warfare is now spreading exponentially with limited controls — like a virus. We see attacks continuing to inflect upward in spite of the race to secure software and the internet, and don't think the potential scale of cyber warfare is well appreciated because of how hard it is to track these attacks systematically."
Frontier labs are not standing still. Anthropic's Glasswing, OpenAI's Daybreak, and Google's Fairwind have driven an unprecedented surge in CVE submissions. The authors call this "meaningful progress toward a more secure internet." But zero-day exploits have roughly doubled since the first truly cyber-capable open-source models were released. Defenders are finding more bugs and attackers are exploiting them faster, simultaneously.
The Patch Window Is Closing
Even when a patch exists, it creates its own attack surface. Fixing open-source software necessarily discloses the underlying vulnerability, and every downstream user must adopt the fix before attackers can reverse-engineer it. AI has compressed the available time. Attar and Habib report that the average window between disclosure and exploitation has fallen from roughly one month to eleven days: "AI has made it easier for attackers to win that race. The number of recorded actively exploited vulnerabilities has surged, and the average time between disclosure and exploitation has fallen from a month to 11 days."
Developers using AI coding tools are simultaneously adding millions of new lines of code at an accelerating rate, each carrying its own potential vulnerabilities. Even if AI-generated code proves marginally safer than human-authored equivalents, the sheer volume means the total attack surface is growing faster than defenders can secure it.
Policy Responses: Directionally Right, Structurally Insufficient
Japan and South Korea have moved with notable speed, with Japan allocating ¥10 billion to a new "top gun" cyber defense team and South Korea mandating executive-level CISOs and steep fines for breaches. In the West, the Trump administration and the UK's Cyber Security and Resilience Bill have added incremental tools. Attar and Habib acknowledge that "some proactive efforts are better than none," but identify three structural gaps none of these measures resolve: the volunteer-maintained open-source software bottleneck; the operational discipline required to adopt patches in under eleven days; and the misalignment risks inherent in AI-powered defensive programs run at speed.
The Window Is Narrowing, and May Not Reopen
The authors are direct: "The 'head start' between Glasswing's launch and the release of cyber-capable open-source models this summer was a one-time gift that we're unlikely to see replicated, and its protective benefit will wane over time. Malicious actors are only starting to learn how to use AI to find and exploit vulnerabilities, and we suspect they are earlier on the learning curve than defenders, with more room to improve."
Fully autonomous attack workflows are not yet broadly deployed, but Bridgewater treats the transition as a near-term planning assumption, not a distant hypothetical.
5 Key Takeaways for Advisors and Investors
- Assign real probability to cyber disruption. One-in-three is not a tail scenario. Pressure-test whether client portfolios are sized for it.
- The asymmetry favors attackers. AI accelerates both sides, but the structural burden falls overwhelmingly on defense. Patch-and-respond strategies are not resolving that imbalance.
- Critical infrastructure exposure is underappreciated. Open-source software maintained by volunteers remains the unaddressed bottleneck, regardless of enterprise cybersecurity spending.
- The exploitation window is now eleven days. Patch-adoption discipline must be recalibrated to a world where known vulnerabilities become active exploits in under two weeks.
- Act before clarity arrives. Attar and Habib warn that waiting for confirmation of damaging attacks "will lead to being too late in markets." Pre-emptive hedging and scenario analysis, not reactive positioning, is the implied prescription.
Footnote:
1 Attar, Hudson, and Arham Habib. "Current Efforts Aren't Enough to Slow the AI Cyber Virus." Bridgewater Associates, 13 Nov. 2026, https://www.bridgewater.com/_document/current-efforts-arent-enough-to-slow-the-ai-cyber-virus?id=000001a1-d6ff-da40-a7e5-f6ff77190000.
AI cyberattack risk, Bridgewater cyber research, AI-enabled cyber warfare, zero-day exploit surge 2026, cybersecurity systemic risk investors, AI cyber arms race financial markets, open-source software vulnerability risk, economically destabilizing cyberattack probability, cyber exploit exponential growth, AI patch window exploitation, financial advisor cybersecurity portfolio risk, Bridgewater one-in-three cyberattack forecast