Do not open a public issue.

Email nejc.cotic@gmail.com with [content-security] in the subject, or use GitHub's private vulnerability reporting.

Please include:

• the version and PHP/Laravel versions
• what an attacker gains
• the smallest reproduction you can manage
• any relevant configuration (policy, driver, disk)

If a proof of concept involves a malicious file, describe it — do not attach it.

You can expect an acknowledgement within 72 hours and an assessment within seven days. Fixes ship as a patch release with an advisory crediting you, unless you would rather not be named.

In scope:

• a bypass that gets a file past a policy that should have rejected it
• HTML that survives the sanitizer and executes in a browser
• a URL the SSRF checks pass that reaches internal infrastructure
• reading or releasing a quarantined file without authorization
• a runtime policy override that widens a control it is documented not to
• the console leaking data to a user the auth callback rejects
• a malformed input that crashes a queue worker or exhausts memory or disk

Out of scope, because the package does not claim them:

• SQL injection that the text scanner did not catch. It is a heuristic signal, not a control. Parameterised queries prevent SQL injection.
• XSS in your own templates. Contextual output escaping prevents XSS. The sanitizer only covers markup you deliberately re-render.
• Malware the engine did not recognise. The package orchestrates ClamAV; it does not write signatures. Missed detections belong upstream.
• Anything with fail_closed turned off, or with the null malware driver in production.
• Findings that require an attacker to already hold admin credentials for the console, unless they escalate beyond what an admin may legitimately do.

These are properties the package intends to hold. A reproducible breach of any of them is a vulnerability, and each has a test:

1. A scanner that cannot answer never produces a clean verdict.
2. A physical storage path is never derived from an uploader-supplied filename.
3. HTML is sanitized by a parser. Never by a regex.
4. Extensions are allowlisted, and forbidden_extensions cannot be widened from the console or the database.
5. The admin console denies by default.
6. End-user validation messages never name the signature or the check.
7. Scanned text and file contents never reach logs, mail, or metadata.
8. Releasing a quarantined file requires a fresh clean scan, or an explicitly requested and audited override.