A critical authentication bypass vulnerability, tracked as CVE-2026-61500, in Rejetto HTTP File Server (HFS) has been actively exploited in the wild, with initial activity originating from a China-hosted IP address. The flaw, which can lead to full administrative access and remote code execution, was discovered by researchers using Anthropic's AI bug-hunting model, Mythos.
Horizon3 researcher Zach Hanley publicly disclosed the vulnerability on Wednesday, October 1, 2026, publishing a video demonstrating the exploit steps. By the following day, exploitation attempts were detected. Patrick Garrity, a security researcher at VulnCheck, reported on Thursday that their canaries detected an actor in China targeting vulnerable HFS hosts in the United States. By Friday, VulnCheck observed four additional hits from two US-based IP addresses, 173.239.211[.]248 and 173.239.211[.]249, which appear to be part of a proxy network.
The vulnerability affects Rejetto HFS, an open-source web file server that was previously listed in the US Cybersecurity and Infrastructure Security Agency’s (CISA) catalog of Known Exploited Vulnerabilities in 2024. Users are advised to update to HFS version 3.2.1 or later to patch this and other security issues.
This marks the second Anthropic-linked vulnerability known to have been exploited in real-world attacks. Mythos and Project Glasswing, Anthropic’s initiative providing select partners access to the bug-hunting model, have collectively identified 286 CVEs since the program's announcement in April. Horizon3 joined Project Glasswing in July and has since used Mythos to uncover numerous critical vulnerabilities.
CVE-2026-61500 highlights Mythos's capabilities in mathematical analysis and computer science. The vulnerability stems from HFS's authentication process, which uses `Math.random()` to generate a value for signing session cookies via the Koa Node.js web framework and its keygrip library. Mythos identified that the V8 JavaScript engine's `Math.random()` implementation, which uses the xorshift128+ algorithm, was not cryptographically secure and its output was fully reversible.
Crucially, Mythos also discovered that the application was leaking `Math.random()` outputs through a separate code path. The AI model recognized that these two facts, when combined, provided sufficient observations to recover the pseudo-random number generator (PRNG) seed. Mythos's analysis further claimed that Z3, a Microsoft-developed Satisfiability Modulo Theories (SMT) solver, could be used to recover the PRNG seed, enabling an attacker to forge valid session cookies and bypass authentication. Researchers noted that the use of an SMT solver in this manner to exploit a cryptographic flaw was an uncommon and sophisticated approach.
