AI models now find software security holes faster than the people who maintain that software can fix them. Here is what the term means, where it came from, who is covering it, and how the next 6 to 12 months could unfold.
The vulnerability tsunami is the idea that AI models now find software security holes faster than the people who maintain that software can fix them. Discovery has become cheap and fast; patching, testing and deploying fixes still runs at human speed. The gap between the two is the tsunami.
The phrase was popularized by Jason Clinton, deputy chief information security officer at Anthropic, in a post on X on Aug 20, 2026. He warned that a wave of AI-found vulnerabilities was months away and that critical infrastructure would not have time to patch, so operators should isolate networks now with air-gaps and data diodes.
The vulnerability tsunami is coming in just a few months and there won't be enough time to patch critical infrastructure. Network isolation described below (air-gaps and data diodes!) needed now!Jason D. Clinton (@JasonDClinton) on X, Aug 20, 2026
On Oct 22, after a week of major incidents, he followed up: things would get worse before they got better, and security leaders should ask their boards for help with the basics, holding on for about 18 months. That post, quoting the August one, drew more than 4 million views and pushed the phrase into mainstream news.
| Date (ET) | Event |
|---|---|
| Aug 20, 4:56 PM | Clinton posts that the vulnerability tsunami is coming in a few months |
| Oct 1, 12:00 PM | Google pauses product submissions to its open-source bug bounty, citing a surge of mostly invalid automated reports |
| Oct 13 | Microsoft Patch Tuesday ships 1,408 CVEs in one day, more than in all of 2025 |
| Oct 21 | PHP2Shell (CVE-2026-109618, CVE-2026-109619) disclosed by Wiz, Calif and OpenAI; PHP ships emergency fixes; CISA urges immediate patching |
| Oct 21 to 22 | Some public services go offline as a precaution, including the Dutch DigiD login service and dozens of UK government agencies, per ExoBrain and The Telegraph |
| Oct 22, 1:43 PM | Clinton: it will get a lot worse before it gets better; hold on for about 18 months |
| Oct 22, evening | Sam Altman: "I no longer think we're going to be able to patch all bugs" |
| Oct 23 | Press wave: The Guardian names the week Black October; Business Insider, Daily Mail, The Telegraph and Anadolu cover the warning |
| Oct 24 | CISA updates its PHP2Shell alert: take unpatchable PHP systems offline |
Links checked Oct 25, 2026.
Vendors are already using the term in their marketing, including ArmorCode (a section titled The Vulnerability Tsunami), Hackuity and Tenable (Vulnami).
| Post | Likes | Reposts | Views |
|---|---|---|---|
| @JasonDClinton, Oct 22: worse before better, hold on ~18 months | 10,010 | 1,343 | 4.3M |
| @JasonDClinton, Aug 20: the vulnerability tsunami is coming | 1,702 | 209 | 979K |
| @DailyMail, Oct 23: story link | 32 | 13 | 9.7K |
| @AGTPinsights, Oct 23: Black October explainer | – | – | 78K |
| @BusinessInsider, Oct 23: story link | 11 | 5 | – |
| Indicator | 2025 | 2026 |
|---|---|---|
| CVEs published, Jan 1 to Oct 13 | 36,998 | 79,680 |
| CISA KEV additions, Jan 1 to Oct 20 | 207 | 270 |
| Average time to exploitation | 72 hours | 24 hours |
| Microsoft CVEs, full year 2025 vs Oct 13, 2026 alone | 1,244 | 1,408 |
| New CVEs received by NVD vs analyzed, 2026 | – | 88,599 vs 41,792 |
Disclosures have roughly doubled while confirmed exploitation rose about 30%: most new CVEs are never exploited, but the ones that are get weaponized faster.
Sources: Empirical Security, Valyu (CISA KEV data), Recorded Future via RECATOOLS, NVD dashboard.
CVE volume stays near 2026's doubled pace. Patch Tuesdays stay huge. A few more PHP2Shell-scale events hit widely used components. Defenders lean on KEV-style prioritization and AI-assisted remediation; most organizations absorb it with longer backlogs and more emergency patch windows.
Attackers automate exploitation as well as discovery. KEV additions accelerate beyond the current 30% growth, edge devices and unmaintained open source become the main targets, and more public services go offline as a precaution. Critical infrastructure without network isolation sees disruptive incidents.
AI labs and vendors ship fixes alongside findings, coordinated disclosure holds, and AI-written patches plus safer-by-default code reduce new bug classes. Disclosure volume stays high but time to patch falls, and the wave peaks within the year.