
If you run IBM's ContextForge MCP Gateway, anything from 1.0.0 to 1.1.0, four CVEs have landed in the last two days and one of them is about replayed signatures. That's CVE-2026-110278. It could let an attacker replay identity propagation claim signatures, or expose sensitive user identity information. In other words, the gateway sees a signature it has already accepted once and accepts it again. The other three: - 110279: the X-Upstream-Authorization header can override the credential - 110195: an authenticated caller can get around upstream credential-level authorisation policies (up to 1.0.11) - 110178: improper access control in the bulk import path for anyone holding a team-scoped credential (up to 1.0.11) Last week it was the SSRF in AWS's mcp-gateway-registry. That makes two gateway vendors and two sets of flaws in one week. I'm not having a go at either team. This stuff is new and it's hard. The replay one is the bit I keep thinking about. A signature only proves something if you can tell it's the first time you've seen it. Otherwise you've got proof of something that happened once being used to justify something happening now. That's why spent-proof replay control has been in AffixIO from the start. It works offline, on your own kit, without calling back to us. Once a proof's been used, it stays used.





