Security

Rapid7 counted AI in 8% of incidents as CrowdStrike rose 6% on Nvidia's agent-security list

Security stocks re-rated on Nvidia's agent-security partner list in the same 48 hours Rapid7's Q3 telemetry put AI inside 8 percent of tracked incidents, with humans still choosing every target. The threat actually on the wire showed up in PHP: two unauthenticated code-execution flaws probed four days before they were disclosed.

Vincent JiangVincent Jiang · 3 min read
Share
Rapid7 Intelligence's Quarterly Threat Landscape Report for Q3 2026, the report that put AI inside 8 percent of tracked incidents
1 / 6Slide 1 of 6
Rapid7 Intelligence's Quarterly Threat Landscape Report, Q3 2026 edition, the telemetry behind the 8 percent AI finding

The partner list moved the money, the telemetry did not

On October 21, CrowdStrike−0.39% — CrowdStrike, down 0.39 percent today rose 6 percent to $241.02 after Nvidia−3.90% — Nvidia, down 3.90 percent today named 11 security partners for its OpenShell agent architecture at GTC Berlin, a list with no dollar terms attached, and the security complex rose as much as 8.3 percent in the session 10. Datadog−3.68% — Datadog, down 3.68 percent today gained 6 percent and Okta−2.32% — Okta, down 2.32 percent today 4 percent, gains one trading desk called expectation-driven until adoption data arrives 9.

The agent-security list re-rated the security complex in one session

0246CrowdStrike6Datadog6Okta4
Data
Value
CrowdStrike6
Datadog6
Okta4
One-day share-price gains on October 21, 2026, the session after Nvidia named 11 security partners for its OpenShell agent architecture at GTC Berlin. Source: 24/7 Wall St.9

The same 48 hours delivered Rapid7's+4.66% — Rapid7, up 4.66 percent today Q3 2026 Quarterly Threat Landscape Report, drawn from the company's own incident-response telemetry: AI was an operational component in 8 percent of tracked incidents, and a human still chose the target in every documented case 15. "The board-deck version of this year has been wall-to-wall AI attackers, and the data doesn't support that story," said Douglas McKee, Rapid7's director of vulnerability intelligence 5.

What attackers actually did in Q3

High-severity vulnerabilities were exploited an average of three days after disclosure, 71 percent inside the first week, and 79 percent needed no authentication and no user interaction, the class Rapid7 calls "holy grail": a door that opens without a login or a click 15. Ransomware stayed human-run at scale, with 1,112 US leak-site listings, roughly eight times Germany's, and two brands in a dead heat 15. High-severity disclosures rose 133 percent year over year, and one in ten public proofs of concept became a working attack within 30 days 2.

Two ransomware brands claimed more incidents than the next three combined

0100200300400The Gentlemen342Qilin339INC Ransom99DeadLock98KryBit95
Data
Value
The Gentlemen342
Qilin339
INC Ransom99
DeadLock98
KryBit95
Claimed ransomware incidents on leak sites, Q3 2026, as tracked in Rapid7's Quarterly Threat Landscape Report.5

PHP showed the real shift: speed, not autonomy

The live counterexample landed the same day. PHP shipped emergency fixes on October 21 for CVE-2026-109618 and CVE-2026-109619, two CVSS 9.2 unauthenticated remote-code-execution flaws in the runtime behind almost 70 percent of websites, and an update is the only complete remediation 378.

Amazon's−4.08% — Amazon, down 4.08 percent today MadPot honeypot logged the first exploitation attempt on October 17, four days before disclosure and while the flaws were under embargo 67. Cloudflare counted thousands of attempts in the 10 hours before the release and millions since, and about one million of 2.6 million internet-facing PHP hosts run versions that no longer get fixes 67.

Amazon CISO CJ Moses named the mechanism: AI-assisted analysis makes it cheaper to turn a patch diff into a working exploit, letting more actors move "within hours or even minutes" 7. Rapid7's own vice president of intelligence, Christiaan Beek, calls the 8 percent "a snapshot of adoption, not a ceiling" 1.

The vendor at the bottom of its range

The company that measured all this closed at $13.22 on October 19, in a 52-week range of $4.97 to $19.84, with insiders net buyers of $3.8 million and a 56-of-100 GuruFocus quality score flagged as a possible value trap 4. Revenue growth has slid from 13.1 percent to negative across twelve quarters 11. The Intelligence unit that shipped same-day detection for GitLab's 10.0 flaw on October 13 10 faces its own test on November 4, when Rapid7 reports third-quarter results 1.

Rapid7's growth slid for twelve straight quarters, into decline

-5%0%5%10%15%Q4 '23Q2 '24Q4 '24Q2 '25Q4 '25Q2 '26Zero growth
Data
Revenue growth, YoY
Q3 '2313.1%
Q4 '2311.3%
Q1 '2412%
Q2 '249.2%
Q3 '248%
Q4 '245.4%
Q1 '252.5%
Q2 '253%
Q3 '251.5%
Q4 '250.5%
Q1 '26-0.3%
Q2 '26-1.5%
Year-over-year revenue growth by quarter, percentages, from Rapid7's SEC filings, retrieved 22 October 2026. The line crosses zero between Q1 and Q2 2026.11

This week's premium accrued to the names attached to the agent story. The attacker already here works in three days and needs no password.

Get Security by email, free

The breaches, flaws and security vendors AI is putting to the test. In our daily email, every weekday.

Loading CRWD…

Reader comments

0 comments

    More in Security

    All in this topic →

    Security

    Fortinet's customers must patch blind: fixes shipped October 22, flaws named November 10, earnings October 28

    Fortinet shipped fixes for five product lines on October 22 and will not name the flaws until November 10, thirteen days after its earnings report. Its own chief information security officer says AI can turn a released patch into a working exploit within hours or days, and that makes the wait itself the exposure.

    Security

    GitLab's second 10.0-severity patch in nine days hits self-managed servers with downtime, not GitLab.com

    GitLab fixed its hosted services centrally on October 21; its self-managed customers face a second upgrade in nine days for a flaw reaching back to 2014, with downtime attached on some servers. The stock trades near its high on AI-driven demand for the governance these flaws keep testing.

    Security

    CrowdStrike rose 6% to $241.02 as security stocks jumped 8% on an Nvidia agent deal with no terms

    Nvidia named 11 security partners for its OpenShell agent architecture at GTC Berlin. CrowdStrike claimed the silicon enforcement layer with no dollar terms attached, and the security complex jumped up to 8.3% in a session on estimates that have barely moved.

    Security

    F5 lands Telenor's sovereign AI contract, then patches three 9.2-severity NGINX flaws

    In one week, Telenor made F5 the service plane of its Nvidia-powered sovereign cloud and F5 rush-patched three 9.2-severity NGINX flaws, on top of a nation-state breach record. The company selling the trust layer of AI clouds keeps supplying evidence against its own premium.

    Security

    Okta's agent gateway still hasn't shipped; rivals enforce its own protocol

    Okta's AI-agent identity premium rests on a product that has not shipped: the Runtime Agent Gateway sits pre-GA while rivals Aembit and TrueFoundry already enforce Okta's own XAA protocol, and the stock has given back a fifth of its October high on no company news.

    Security

    Check Point found LiteLLM let any signed-in user seize admin; 97 million downloads a month

    Check Point showed that any signed-in LiteLLM user could seize the admin account of the gateway holding a company's every model key, the latest receipt in a year of backdoors and botnets. The biggest customers are now rewriting the layer rather than upgrading it, and the free middle of the AI stack is becoming paid infrastructure.

    Get the newsletter

    The day's AI and market news, in your inbox every weekday.