CVE-2026-73603

Flowise before 3.1.4 contains a missing-authorization flaw (CWE-862) in its unauthenticated text-to-speech endpoint, which does not check a chatflow's visibility setting before processing requests. An unauthenticated remote attacker who supplies a valid chatflow UUID to the TTS endpoint can make the server generate audio using the chatflow owner's stored OpenAI or ElevenLabs API keys. The attacker gains free, unlimited text-to-speech generation billed to the victim's provider accounts, while chatflow data itself is not exposed (CVSS impact is limited to low integrity/availability, and this is a cost-abuse issue rather than data theft). Any Flowise deployment running a version before 3.1.4 is affected, with the highest practical risk on internet-exposed self-hosted instances that have configured TTS provider keys. No in-the-wild exploitation is currently documented: a public proof of concept is available via the vendor security advisory, and EPSS assigns only a 0.3% probability of exploitation in the next 30 days.

What to do: Upgrade Flowise to version 3.1.4 or later. As interim mitigation, restrict unauthenticated access to the TTS endpoint (e.g., reverse-proxy authentication or network rules) on internet-exposed instances, and monitor/rotate stored OpenAI and ElevenLabs API keys for unexpected usage. Check access logs for unauthenticated TTS requests referencing chatflow UUIDs.