Fortinet published details of nearly 100 vulnerabilities in its security products Tuesday, about 17 of them rated critical, SecurityWeek reported Wednesday. Two of the most serious were found by Fortinet's own engineers using Anthropic's Claude Mythos AI model, according to the company's advisories, and the new details give attackers a guide to any Fortinet equipment that has not yet been upgraded.

Fortinet has not said that any of the flaws are being exploited, SecurityWeek noted. The fixed software has been available since October, when Fortinet asked customers to upgrade before it described the bugs.

What Happened

Fortinet released 89 advisories covering nearly 100 flaws, about 17 critical and nearly 80 rated high severity, according to SecurityWeek. FortiMail, the company's email security gateway, received the largest share of critical fixes, followed by FortiAuthenticator, the FortiOS firewall operating system, the FortiManager management console and the FortiAnalyzer logging product. FortiSASE, Fortinet's cloud security service, and the FortiClient endpoint software also got critical fixes. The possible consequences include running attacker code, bypassing sign-in, taking over accounts and executing commands as a super administrator.

RelatedMicrosoft Researchers Plant Sleeper Backdoor in Open-Weight AI Models by Editing Under 10 Neurons

Two advisories stand out. One, rated 9.0 out of 10, affects FGFM, the protocol FortiManager uses to talk to the devices it manages. Fortinet's advisory says an attacker who can reach the FGFM port could run code or commands with crafted network traffic, without logging in. Fortinet lists no workaround.

The other, rated 9.3, sits in the part of FortiOS and FortiSASE that handles OSPF, a protocol routers use to share maps of a network. Fortinet says an unauthenticated attacker could use a crafted request to broadcast routing announcements. FortiOS is fixed in versions 8.0.2, 7.6.8, 7.4.13 and 7.2.14, the builds Fortinet released in October. Several FortiSASE branches cannot be patched in place; Fortinet tells customers on them to migrate to a fixed release.

Both advisories credit members of Fortinet's Product Security team working "with the use of the frontier AI model: Claude Mythos." A third critical advisory from the same day covers an authorization bypass in FortiManager and FortiAnalyzer.

Outside researchers also contributed. Kevin Joensen of Baldur Security said, according to SecurityWeek, that a FortiClient flaw he calls "Fortinet2Shell" could let a malicious website run code on a computer when the FortiPAM feature is enabled. He traced it to a misconfigured cross-origin resource sharing (CORS) setting, the web rule that decides which sites may talk to a service, which let attacker-controlled pages reach a service on the machine.

Context

Help Net Security reported Oct. 26 that Fortinet shipped the fixed builds the previous week and held the details until Nov. 10. Fortinet Chief Information Security Officer Carl Windsor said the flaws were found with AI models and confirmed by people. The Primary Wire reported at the time that the company was asking customers to patch before it could say what it was fixing.

In the new release, Fortinet said the delay gave customers time to upgrade before details became public, which it said "matters even more as AI shortens the time between disclosure and exploitation," according to SecurityWeek.

Why It Matters

Fortinet firewalls and consoles sit at the edge of corporate networks and are frequent targets. The Primary Wire reported in October, citing BleepingComputer, that attackers exploited a critical FortiMail flaw before a fix existed. Weeks later, CISA confirmed attacks on another Fortinet flaw that researchers tied to the theft of data from more than 500 firewalls, showing how one compromised management console can expose many customers.

For executives, the question is whether October's upgrades were actually installed. Security teams should confirm the fixed versions on every FortiOS firewall and on FortiManager, FortiAnalyzer, FortiMail and FortiAuthenticator, make sure FortiManager's device-management port is reachable only from the equipment it manages, and check whether any FortiSASE deployments need migrating. Organizations using FortiClient with FortiPAM should treat that update as urgent.

The release also shows a pattern: vendors pointing frontier AI models at their own code are publishing far larger batches of fixes at once, which puts more pressure on patch teams.

What to Watch

Watch for public proof-of-concept exploit code now that details are out, and for any of the flaws to appear in CISA's catalog of actively exploited vulnerabilities. Also watch whether Fortinet updates the advisories to report attacks, and whether other vendors disclose similar AI-found batches.