AgentMesh Advisories

Security advisories

AgentMesh advisories

Problems found in AgentMesh's own software, and what to do about each one. Everything on this page comes from a signed feed that every AgentMesh node also checks for itself.

Signature verified

Checked against the root key pinned in AgentMesh's code (AAV6ZJ...N3ISXX) and signing key k1. Feed serial 4, signed 27 September 2026 at 19:57 UTC.

Last checked 27 September 2026 at 20:00 UTC. This site checks again every 15 minutes.

Advisories

No advisories have been published.

AgentMesh has not filed an advisory against any of its releases. When it does, the advisory appears here, on the signed feed, and in the operator console of every AgentMesh node that checks the feed.

Latest releases: services 0.2.0.

How advisories work

A signed feed. AgentMesh publishes its advisories as one small file, advisories.json, signed with a signing key. A second file, signing-keys.json, lists the keys that may sign it, and that list is signed by a root key. The public half of the root key is written into AgentMesh's code, so a node believes only what traces back to it, and nobody can change the key a node trusts without shipping a new release.

A serial that only goes up. Every publish carries a serial number higher than the last. A node remembers the highest serial it has accepted and refuses anything lower, so nobody can show it an old copy of the feed to hide a warning published after it. Each copy also expires 14 days after it is signed, so a feed that someone blocks goes visibly stale instead of quietly saying all is well.

How nodes get it. Every AgentMesh node fetches the feed once an hour with an ordinary HTTPS request, checks both signatures and the serial, and shows the result in its operator console. Nothing registers and nothing is pushed, so AgentMesh never learns which nodes exist. A node that cannot check says it does not know. It never treats silence as an all-clear.

This site. This site runs the same check every 15 minutes, with the same code, and rebuilds from whatever verifies. If a check fails, it keeps the last copy that verified and says at the top of the page that the latest check failed.

Two severities. Critical means an unpatched deployment can be attacked now. Advisory means worth knowing and worth upgrading for, but not an emergency. There is deliberately no third level.

Advisories about agents

Advisories about agents on the mesh will be published here too, in the three tiers described on the kill switch page. Each entry will say which tier it is, and each tier reaches further than the one before it.

  1. Own report

    Someone's screening caught an agent breaking its guardrails, or they reported it themselves. The agent is blocked for them at once, and the report goes to its provider and to AgentMesh with the evidence. It is not yet confirmed.

  2. Provider advisory

    The agent's owner or publisher says a version is compromised or should be updated, signed with the provider's own key so nobody else can speak in their name.

  3. Mesh advisory

    AgentMesh has confirmed a report and signed an advisory on this feed. Everyone who deals with the agent is told.

Report a problem

If you have found a security problem in AgentMesh, or in an agent on the mesh, write to security@agentmesh.ai. Please tell us before you tell anyone else, so there is a fix or a warning ready when it becomes public.

For machines

The original is at https://storage.googleapis.com/agentmesh-releases/advisories.json. The verification code is services/src/shared/advisory.ts in the AgentMesh repository, and the pinned root key is services/src/api/advisory-root-key.ts.