ARTEX is an autonomous penetration-testing tool. Security fixes are provided only for the latest official release line. The main branch is not considered a stable release and may change at any time.
Only the latest release line receives security fixes. Older releases do not receive fixes; please upgrade to the latest release.
Notes:
- Please upgrade to the latest release before reporting a vulnerability.
- Forks, custom builds from arbitrary commits, and modified versions are not covered by official support.
Do not open a public issue for a security vulnerability.
Sign in to GitHub and use the private vulnerability report form. You can also reach it from the repository's Security and quality tab under Advisories → Report a vulnerability. This is the only reporting channel.
Please include:
- A description of the vulnerability and its expected impact.
- The ARTEX version and/or commit hash.
- Runtime environment (OS, Docker/local, PostgreSQL, LLM provider, etc.).
- Minimal reproduction steps (PoC).
- Logs, screenshots, or sample traffic with sensitive data removed.
- Suggested remediation, if available.
Do not send real third-party data, credentials, tokens, customer data, or unredacted sensitive content.
ARTEX is an offensive-security tool. Test only against:
- Your own ARTEX deployment in a lab or local environment;
- Systems you own;
- Targets explicitly authorized in writing (bug bounty, pentest contract, etc.).
Testing, scanning, or exploiting third-party systems without explicit authorization is strictly prohibited. Reports that violate this principle may be closed without action.
ARTEX trusts the scanner output file (output.json) that ingest_bbot reads. The scope_distance field carried on scope-relevant events is the scanner's own verdict and is ingested as-is: an event the scanner marks in scope (per its own -s/--no-scope configuration) is treated as in scope without being re-checked against ARTEX's rule set.
The enforced trust boundary for asset insertion is insert_assets, which runs every candidate through EvaluateAssetGate: block rules always apply, and a task allow-list is enforced when one is configured. A rejected candidate is returned as an insertion forbidden error for that asset and is never inserted, so it cannot reach task execution. in_scope_not_asset in the ingest_bbot result is a different count: in-scope events that are not asset types (findings and the like).
Accepted gap: when a task has no allow-list, there is no allow rule to apply (block rules still run), and any host carried on an in-scope event that is not blocked enters the graph as an asset. Operators who rely on the scanner's scope rules alone should configure an allow-list per task, or review output.json before feeding it to ARTEX. Cross-checking ingested events against ARTEX's own rule set would change what existing scanner configurations mean for those operators, so it is treated as a deliberate design question (see #317), not an accepted vulnerability.
POST /api/llm/models fetches model lists from the provider endpoint and optional HTTP proxy supplied in the request. By default, ARTEX allows only http and https URLs whose hosts resolve exclusively to public IP addresses (blocking loopback, RFC1918 private networks, IPv6 unique-local fc00::/7, link-local/cloud metadata such as 169.254.169.254, unspecified, and multicast addresses), and enforces the same checks on redirects and at dial time. The built-in local Ollama default (http://127.0.0.1:11434/v1) is permitted only when the request leaves base_url empty for the ollama provider.
Operators running self-hosted LLM gateways or Ollama on a LAN or Docker bridge can allow specific endpoints with ARTEX_LLM_PRIVATE_ENDPOINTS, a comma-separated list of exact host:port entries (for example, 127.0.0.1:11434,host.docker.internal:11434,192.168.1.10:11434). A URL whose host:port matches an entry may resolve to a loopback or private address while scheme restrictions, link-local/cloud-metadata blocks, and redirect validation still apply; everything else stays refused. Leaving ARTEX_LLM_PRIVATE_ENDPOINTS unset or empty keeps the strict public-only policy.
The server runs as root, but the processes an agent starts (Bash, the interactive shell, custom tools and stdio MCP servers) run as the unprivileged uid and gid set by -exec-uid and -exec-gid (default 10001). Only the run directories and their own home (<data>/exec-home) are writable by them. No agent working directory is an ancestor of another: the main agent and the planner work in tasks/<id>/main, a worker in tasks/<id>/i<N>, a chat in sessions/<id>. Their parents (tasks/, sessions/ and tasks/<id>) are root:root 0755, so the executor cannot create, rename or remove an entry there, and only the leaves are shared with the executor group (setgid, 0775); with the drop active the server creates each level without following a symlink. Entries that a task already had directly under tasks/<id>/ before this layout (uploads, files the main agent wrote there) are not moved and are shared with the executor group again, at start-up and on archive restore, so for tasks in progress too, while tasks/<id> itself stays root:root 0755: the main agent's new run directory main/ is empty on its first turn, and the earlier artifacts stay under tasks/<id>/ and are reached through Bash only. keys/ (jwt.key, secret.key), the config file and the MITM CA key are root-only, and the server's environment block no longer carries ARTEX_PG_DSN, ARTEX_SECRET_KEY, the LLM API keys or the LLM proxy and base URLs (Linux). The agent's file tools (Read, Write, Edit, MultiEdit, LS, Glob, Grep) run inside the server but reach only the run's working directory: paths are opened through os.Root, so absolute paths elsewhere, .. and symlinks leading out are refused, and a working directory reached through any symlink is refused. The executor cannot replace a run directory or one of its parents, so the race that remains is the swap between an EvalSymlinks check and the open of a file inside a run directory, which the executor can do only inside a leaf (every executor process has the same identity, so this is any leaf, never a leaf itself or its parents). ingest_bbot reads its file through the same confinement. Custom-tool scripts are protected only when the executor privilege drop is active: they are then written to <data>/exec-scripts, a root-owned directory the executor can read and cannot change; with the drop disabled they still go to .tools/ under the run directory. Command-output spills (the full output of a tool call over the size cap, written by Norma inside the server) follow the same rule: with the drop active they go to a root-owned directory per task or chat session outside the run directory, <data>/spill/tasks/<id> or <data>/spill/sessions/<id> (root with the executor group, 0750, below root:root 0755 parents), which is deleted and archived with the task, or deleted with the conversation. The executor cannot create, rename or remove an entry there. Norma writes a spill file 0600 and the server makes it root with the executor group, 0640, when the tool call returns: the agent reads it with Bash, not with the file tools, which do not reach outside the run directory. With the drop disabled spills still go to cmd-output/ under the run directory. Residual, not closed here: the other root reads and writes by path remain open, and a symlink the executor places can redirect them, so root may read or write outside the workspace. They are ingest_scan_output (a root read by path), chat uploads under the tasks/<id>/main/uploads/ or sessions/<id>/uploads/ directories, task archive packaging (lstat then open by path), and the background-task output that Norma writes inside the server under os.TempDir()/norma (<tmp>/norma/<session>/tasks/), a path ARTEX does not choose: with the drop active the server creates that directory root:root 0755 at start-up, before any agent process runs, and refuses to start when it is a symlink or not a directory, or when a pre-existing <tmp>/norma is one the executor can write (this can refuse a start that worked before; the operator remedy is to remove that directory or set TMPDIR to a directory the executor cannot write; ARTEX never modifies or deletes the directory), but a directory already there under another owner is used as found, with a warning in the log. TMPDIR must be a root-controlled path: the start-up check follows TMPDIR and inspects <tmp>/norma and its immediate parent only, so a symlink or an ancestor below a directory the executor can write is not detected, and POSIX ACLs are not inspected. The ask_operator image staging is narrower than those. It creates tasks/<id>/uploads with Mkdir and checks it with Lstat (a symlink or anything that is not a directory is refused), gives the copy a random name ask-operator-<unixnano>-<8 hex>.<ext> created with O_EXCL and mode 0640, so it never overwrites a file, and opens the source only when the Lstat of the path and the Stat of the opened file agree (os.SameFile). What remains there: a directory component inside the run directory can be swapped for a symlink between the check and the open, so root can still read a file outside the run directory and stage a copy of it; an uploads/ directory that existed before start-up or an archive restore is shared with the executor group and stays group-writable, so a staged copy can be replaced after it is written and the operator may see another image; with fs.protected_hardlinks=0 a hard link to a file the executor cannot read is copied; and a FIFO swapped in after the Lstat blocks the call. The Docker image has no setuid programs and docker-compose.yml sets no-new-privileges. These are a containment layer, not a sandbox: an executor can still read world-readable files under the data directory, reach the loopback API and the database if the network allows it, and use any credential it was given through the platform. Persistence tools (custom tools, MCP servers, skills) need operator approval by default.
This is an open-source project maintained on a best-effort basis. We aim to acknowledge a report within about a week and keep you updated, but we cannot guarantee specific response or fix timeframes.
- We will open a private GitHub Security Advisory for coordination.
- We will identify the release containing the fix when it is published.
- We may request a CVE if the vulnerability qualifies.
- We will credit you or your organization in the advisory if you wish.
- Default disclosure timing: 90 days after a fix is available, or earlier by mutual agreement. If the vulnerability is being exploited in the wild, we may disclose earlier.
We will explain the reason, for example:
- It is not a security vulnerability but intended behavior.
- It is outside ARTEX’s scope (e.g., an issue in an LLM provider, PostgreSQL, Docker, or a browser).
- There is insufficient information to reproduce it.
- The report only states that “the tool can be misused” without identifying a specific technical flaw in ARTEX.
Where appropriate, we may convert the report into a public hardening issue or feature request.
The following are generally not considered ARTEX security vulnerabilities:
- Attacks or scans against systems you do not own or are not authorized to test.
- Self-XSS, spam, or brute-force without real impact.
- Large-scale DoS/DDoS or resource-exhaustion attacks.
- Social engineering targeting maintainers or users.
- Issues that exist only in third-party dependencies and do not affect ARTEX.
- Reports that “ARTEX can be used to attack systems” when the user has valid authorization — this is the nature of a penetration-testing tool, not a vulnerability.
- Missing headers or hardening configuration without a concrete security impact.
We welcome ethical security research. If you follow this policy and test only within authorized scope, we will not pursue legal action against you for good-faith research. This does not exempt you from applicable laws in your jurisdiction.
- Do not publicly disclose vulnerability details before a fix is available or before the coordinated disclosure period ends.
- Do not publish a public PoC without maintainer consent.
- If you do not receive a response within a reasonable time, please ping the contact channel above before disclosing.
Thank you for helping make ARTEX safer.