On this page (7 sections)
A critical flaw in Atomic Server, an open source headless CMS and real-time database written in Rust, lets an unauthenticated visitor take over a freshly deployed instance. The server creates a well-known invite at /setup, and anyone who can reach that URL can use it to create a new Agent with write access to the entire Drive. The issue is tracked as CVE-2026-69157 and rated 9.8 (critical).
If you run Atomic Server, especially a new instance that you have not yet claimed, or one that is reachable from the internet, you should read this today. At the time of writing, the GitHub advisory lists no patched version, so the practical work is reducing exposure rather than simply upgrading.
What happened
According to the GitHub advisory, Atomic Server provisions an invite at the /setup path with write permission enabled, the Drive root as its target, and a single allowed use. The code that handles invite redemption discards the identity of whoever redeems it, so holding the URL is the only requirement. There is no password, token or operator login in the way.
A visitor supplies a public key of their own choosing. The server then creates a new Agent for that key and adds it to the Drive's write rights. On a fresh deployment that has not been claimed by its legitimate owner yet, the advisory calls this a full server takeover with no credentials.
The advisory was published to the repository on Jul 6, 2026, reached the National Vulnerability Database on Nov 17, 2026, and was reviewed in the GitHub Advisory Database on Nov 18, 2026. A fix commit is referenced, but the advisory still shows no patched release.
Why it is dangerous
This is a missing-authentication problem. The setup invite exists so the first operator can claim a new server, but it is guarded only by being obscure and well known at the same time: the path is the same on every install. Anyone who finds an Atomic Server on the network can ask for it.
With Drive-wide write access, the advisory says an attacker can:
- read, modify or delete any resource in the Drive
- create or alter Agents
- change permissions
The advisory also says this serves as the entry point of a larger chain: with write access, an attacker can read the server root private key and impersonate root. That follow-on issue is described as reported separately, and the research gives no further detail, so treat it as a reason to assume full compromise rather than a documented second CVE.
Related weaknesses noted in the advisory
The advisory's remediation section points at two further defects in the same invite code. First, the single-use counter is checked and decremented in separate steps, which the advisory describes as a race that can let a one-use invite be redeemed several times at once. Second, the expiry check is described as inverted, so expired invites stay redeemable while valid ones are rejected. These come from the advisory's own analysis; the research does not show independent confirmation.
Who is affected
The advisory lists atomic-server versions from 0.29.0 up to and including 0.40.1 as affected. There is a discrepancy to be aware of: the CVE description text says the range runs from 0.29.0 to 0.40.2, while the structured package range and the advisory say <= 0.40.1. The research does not resolve this, so if you run 0.40.2, do not assume you are safe. The same CVE record states that no complete fixed version was verified.
Realistic exposure is highest for:
- new deployments where nobody has yet claimed the server through the setup flow
- instances reachable from the public internet or a shared network, including test and staging copies
- containers started from a development image and left running with the default port published
The advisory frames the full takeover as applying to a fresh, unclaimed instance. It does not say how an already-claimed server behaves after the invite has been used, and the race condition it describes suggests you should not assume a used invite is safely closed. Verify rather than assume.
How to check your exposure
You can work this out without touching the vulnerable endpoint yourself. Start with inventory, then review who holds write rights.
- List every Atomic Server instance you run, including staging, demo and forgotten containers.
- Record the version of each one and compare it with the range 0.29.0 to 0.40.1 (and treat 0.40.2 with caution given the conflicting record).
- Check whether each instance is reachable from outside your trusted network, for example by looking at published container ports, firewall rules and reverse proxy configuration.
- Open the Drive root on instances you own and review the list of Agents in its write rights. Any Agent you do not recognize means you should treat the instance as compromised.
- Review reverse proxy or access logs for requests to /setup, especially ones with a public-key query parameter, from addresses you do not recognize.
This command searches an nginx or Apache access log for requests to the /setup path. It is read-only; run it on the proxy host and look for any hits you cannot explain. Adjust the log path to your setup.
grep -F "/setup" /var/log/nginx/access.logWhat to do now
Because no patched release is listed, the goal is to stop untrusted clients from reaching /setup and to confirm nobody has already used it.
- Claim or remove unclaimed instances first. A server nobody has set up yet is the exact target; do not leave one running on a reachable address.
- Restrict network access. Put Atomic Server behind a VPN, a private network or an IP allowlist so that only trusted clients can connect, until a fixed release is confirmed.
- Block /setup at your reverse proxy or firewall for everyone except an administrator address, once your own setup is finished.
- Audit Agents and write rights on the Drive root. Remove any Agent you did not create, then review resources for unexpected edits or deletions.
- If an unknown Agent was present, assume the server root private key may have been exposed, as the advisory describes for this attack chain. Rebuild the instance from a known-good backup and issue new keys rather than trying to clean it in place.
- Watch the project's advisory for a patched release, and upgrade as soon as one is published and verified.
What the fix should look like
The advisory's remediation guidance describes what a correct fix needs: a one-time secret printed at startup, or operator authentication, required to redeem /setup; the invite invalidated after first use through an atomic decrement rather than a read-then-write check; and the inverted expiry comparison corrected. When a new release appears, check its release notes against those three points before you treat the problem as closed. The research does not confirm that the referenced commit delivers all of them, which matches the CVE record's statement that no complete fix was verified.
Detection and monitoring afterwards
- Keep an alert on any change to the Agent list or write rights on the Drive root.
- Log and review requests to /setup on any proxy in front of the server.
- Track the advisory and the repository for a patched release, since the status could change after this article was written.
- Monitor the certificate, DNS and response headers of the public hostname so that unexpected changes to an exposed admin-style service are noticed quickly. Webforta Monitor covers certificate, DNS and header-change monitoring, though it does not detect this specific flaw.
To see what your public hostnames expose, a read-only check with Webforta Scan of a site you have verified can help you review your external surface. For general guidance after a suspected takeover, see website hacked: what to do.
Sources
This article is based on the public sources below. Details can change as vendors and researchers publish more; check the original advisories before acting on version numbers.
