Fortinet Patches 100 Flaws, 17 Critical, Found With Frontier AI Models

Fortinet disclosed nearly 100 vulnerabilities across its product line on November 10, and about 17 of them are critical. SecurityWeek reported that nearly 80 of the rest are high-severity, and Fortinet has not said any are being exploited in the wild.

The batch is unusual for its origin. Fortinet says some of the flaws were surfaced by frontier AI models it has folded into its product security testing program. The company publicly credited four partners: "Fortinet would like to thank Anthropic, OpenAI, Google, and SpaceX for their collaboration throughout this process." Anthropic, OpenAI and Google supply frontier models; SpaceX is an odd name in that list, and Fortinet did not explain its role.

According to Fortinet, the models "analyze code paths, protocol handling, and configuration logic at a depth and speed that traditional fuzzing and manual review cannot match." That claim is a roadmap dressed as a result, but the patch count is a confirmed figure, and it is large by any vendor's monthly standard.

Which products carry the worst bugs?

FortiMail took the largest share of critical fixes, followed by FortiAuthenticator, FortiOS, FortiManager and FortiAnalyzer. Updates for FortiSASE and FortiClient also resolved critical issues.

Two FortiMail flaws stand out. A path traversal bug and a command injection flaw reachable through crafted HA join-request packets both carry a CVSS score of 9.8, and both let an unauthenticated attacker execute code or commands. Other FortiMail fixes address takeover of IBE recipient accounts and hijacking of encrypted webmail sessions.

FortiManager carries an improper-authorization flaw in its FGFM protocol. An unauthenticated attacker who can reach the FGFM port can run code on the device.

FortiClient contributes its own entry. Baldur Security's Kevin Joensen named the bug Fortinet2Shell, and he says it stems from a CORS misconfiguration. A malicious web page can reach a local service and "reliably execute arbitrary code" when FortiPAM is enabled.

Why the urgency language?

Fortinet told customers to treat the fixes as "a priority change rather than a routine maintenance cycle." The company's reasoning cuts both ways. AI now helps attackers diff patches into working exploits "in hours or days, rather than weeks," Fortinet warned, and it can chain lower-severity bugs into something worse.

The same October timing matters here. Fortinet shipped a coordinated "security hardening release" that month, pushing fixed builds of FortiMail and FortiAuthenticator first, then FortiOS, FortiManager and FortiAnalyzer on October 22. It withheld the details until this month's Patch Tuesday. Customers who patched early were protected before the advisories went public; those who waited now face published vulnerability descriptions with working exploits plausibly close behind.

One clarification for anyone tracking Fortinet advisories: this batch is separate from CVE-2026-104286, a FortiMail flaw that CISA and Fortinet flagged as actively exploited in early October. That one already has a place in CISA's catalog.

How does this fit the month's wider picture?

The volume follows a pattern. Microsoft's November Patch Tuesday fixed 1,093 flaws, including an exploited Outlook bug credited to a researcher using Anthropic's Claude. SAP shipped four maximum-severity fixes in the same window.

The symmetry is hard to miss. The same class of models now finds vulnerabilities on the defense side and accelerates exploitation on the attack side, and neither side has a durable advantage. Fortinet's move to institutionalize AI-assisted testing suggests the company expects the discovery rate to keep rising — and customers should budget their patching cycles accordingly.