Skip to content

Latest commit

 

History

History
121 lines (103 loc) · 6.46 KB

SECURITY.md

File metadata and controls

121 lines (103 loc) · 6.46 KB

Reporting a vulnerability

Do not open a public issue for a security problem.

Use GitHub's private vulnerability reporting: Report a vulnerability

That channel is private between you and the maintainers until an advisory is published. It needs no email address from either side, which is deliberate: a disclosure process should not depend on anybody publishing a personal address. Today there is one maintainer, Rashik Adhikari (@rsh1k), who receives every report.

If private reporting is unavailable to you for any reason, open a public issue containing only the words "security contact requested" and nothing else, and a maintainer will open a private advisory to continue in.

What to expect

First response within 5 days
Assessment and severity within 14 days
Fix for a confirmed high or critical issue within 30 days, or a written explanation of why longer
Credit in the advisory, under whatever name you choose, including none

There is no bounty. This project has no money. What it can offer is that your report is taken seriously, fixed properly rather than patched around, and credited.

What counts

Aertun makes specific structural claims, and a demonstration that any of them is false is a vulnerability even without a working exploit:

  • Authorisation is enforced at the action, not the screen. Any request that performs an action the caller's role, job or permission does not allow, through the console, the command line or the agent interface — and any state of the store in which the checks stop happening at all. A damaged policy file must refuse, not permit.
  • A credential is capped by what it carries. Any way a token acts beyond its role, the area it was issued for, or its read-only limit, or a session outlives the limits it was issued under.
  • Two factors for everybody. Any way a person reaches the console on one factor once two are required, or single sign-on counts as two factors when the provider did not say it used two.
  • Nobody acts alone where two are required. Any way an act on another platform is approved by whoever asked for it; a run an automation held is approved by whoever wrote the rule; or somebody is made an administrator, an administrator's token is issued, a proof signer is trusted, a setting is weakened or the audit log is moved on one administrator's word while another exists; or the only administrator's own request is approved without its reason, before its delay, or without a passkey sign-in from the last five minutes.
  • The Owner cannot be removed. Any way another administrator, single sign-on provisioning, a deny or an end date takes the Owner's administration of the store away, or ownership moves without the Owner asking and confirming it a day later.
  • A run token cannot be widened by whoever holds it. Any way to remove, widen, add or reorder a caveat of an agent's run token, or move it to another run, and still have the tool gate accept it without the minting key.
  • Failures lock out places, not people. Any way failed sign-ins from somewhere else lock a person out of the places they sign in from.
  • The audit log is append-only and verifiable. Any way to alter, drop or add a record and still have aertun auditlog verify pass, a signed tree head verify over a log it does not cover, or a streamed batch be accepted by a receiver that checks it with a record changed or missing.
  • An agent reads and proposes, and does nothing else. Any way an agent connected over MCP records a decision, approves an act, changes who may do what, or reaches another platform.
  • Its own proofs say what was checked. Any signed proof that verifies while saying something its evidence does not support, or that carries text an attacker could have written.
  • What an agent reads cannot instruct it. Any way text in an event, a finding or a log reaches the analyst model as anything but fenced data, or the analyst records a decision rather than a proposal.
  • Personal data stays out of prompts. Any way a value the privacy guard masks reaches a model that may not receive it.
  • A credential sealed is a credential kept. Any way a connector's credential or a feed's secret is stored in the clear once a keyring exists, or read with a key other than the one it was sealed under.
  • A connector reaches only what it declares. Any request a connector or a feed makes to a host its manifest does not name, or to this machine's private network when it was not allowed to.

What does not count

  • Findings against a deployment that has the console on a public interface without TLS in front of it. aertun serve listens on loopback by default.
  • Missing hardening headers on a response that carries no content.
  • Denial of service by sending very large or very many requests.
  • Automated scanner output with no demonstrated impact.
  • Anything requiring an already-compromised host or an already-valid administrator credential.

Supported versions

Version Security fixes
0.1.x yes
main yes: fixes land here first

Report against the latest 0.1 release or against main.

Our own claims, tested

The properties above are asserted by the test suite rather than argued for, and a report is most useful when it makes one of those tests fail:

  • internal/auth — roles, jobs, permissions, scopes and sessions
  • internal/admin/roles_test.go, internal/admin/jobs_test.go — every role and job reaches its own work and no more
  • cmd/aertun/privilege_test.go, cmd/aertun/coverage_test.go — every command declares what it needs, and has a screen and an MCP operation or a written reason
  • internal/admin/mfa_test.go, internal/admin/stepup_test.go, internal/admin/sessions_test.go — two factors, step-up and session limits
  • internal/lockout — smart lockout
  • internal/dual, internal/incident, internal/automate, cmd/aertun/soloreason_test.go, cmd/aertun/ownercmd_test.go — second people, the Owner, and approving alone
  • internal/macaroon, cmd/aertun/caveat_test.go — run tokens and their caveats
  • internal/proof, cmd/aertun/proofcmd_test.go — signed proofs
  • internal/audit, internal/translog, internal/auditstream — the chain, the tree, signed heads and the stream
  • internal/pii, internal/model — the privacy guard in front of a model
  • internal/fetch, internal/egress, internal/connector — what may be reached