locklog lets an AI agent search and count across logs while a console held by a human reads them: the personal data and secrets locklog detects are masked before anything reaches the agent. locklog is pre-release (0.x), early and in active development. The console runs in Docker, in a container and under an account of its own, which the agent must not be able to reach (a deployment precondition, P1 in docs/security-model.md). A flaw in the masking, the console or its socket is a security issue.
Please report vulnerabilities privately through GitHub security advisories: open https://github.com/DavidGodefroid/locklog/security/advisories/new (the "Report a vulnerability" button on the repository's Security tab).
Do not open a public issue, pull request or discussion for a suspected vulnerability.
Please include:
- the locklog version (
locklog version: the launcher's and the image's) and your OS and Docker version; - the log source (files, journald, containers, Loki, ...) and how the agent reaches the console, if relevant;
- the steps or input needed to reproduce the issue, and its impact.
Never include real credentials, tokens, cookies, real personal data or unmasked log lines in a report: use synthetic values.
You will get an acknowledgement within a week. Fixes are released as soon as possible, and the advisory is published with credit to the reporter unless you ask otherwise.
Issues of interest include, among others:
- a value of a kind locklog detects (an email address, a phone number, an IBAN, a card number, a national id, a public IP address, a secret of a known shape) reaching the agent, the audit log or an error message unmasked;
- a way for the agent to learn a masked value without a human's approval: through search predicates, counts, patterns, result sizes, cursors, stopping points or timing;
- a way for the agent to read a log source the human did not approve, to write to any log source or backend, or to loosen the policy without the human's confirmation;
- a way for the agent to reach the console's account, its home volume or a backend credential;
- log content that can spoof the console's approval screen or make locklog act beyond the agent's call.
What locklog does not detect is listed in docs/security-model.md and is not a vulnerability by itself. Attacks that require running code as the console's OS account, or an agent that can already read the logs, the Docker socket or the backends directly (a deployment precondition), are out of scope.
Releases before v1.0.0 are early releases: their interfaces and formats can change between minor versions. Only the latest release receives security fixes.