The latest released minor version receives security fixes.
Please report suspected vulnerabilities privately (for example through GitHub's private vulnerability reporting) rather than in a public issue. Include a description, the affected version, and reproduction steps where possible.
In scope:
- Leakage of the bot token or the webhook secret.
- Privilege escalation: a non-administrator executing an administrative command, or acting on a protected target (the owner, another administrator, or the bot).
- Forged webhook updates or captcha callback clicks.
- Parser bugs reachable from untrusted input (update JSON, callback data, blacklist patterns).
- Denial of service through a hostile blacklist regular expression.
- The token and webhook secret are redacted from logs and diagnostic output. The
redaction runs on the final rendered log line, so it also covers a URL
assembled with
%-style arguments, the formatted exception andstack_infotext, and every logger that propagates to the root handler (aiogram,aiohttp,asyncio). Registered values (the configured token and secret, the URL-quoted form, and the part of a token after the:) are matched exactly, and token-shaped strings (bot<digits>:<secret>, bare<digits>:<secret>) are matched by pattern, so an unregistered token is caught too.Settingsmasks its secrets inrepr/str, and a failed config validation no longer echoes the value it rejected. - Webhook updates are authenticated with a constant-time comparison of the
X-Telegram-Bot-Api-Secret-Tokenheader; a mismatch is rejected with 403, and the bot refuses to start in webhook mode without a secret of at least 16 characters. - A config file that holds a secret must be private: if it has any group/other
permission bit set the bot refuses to start and asks for
chmod 600. The check resolves symlinks (the target's mode is used).--allow-insecure-config(envALLOW_INSECURE_CONFIG=1) exists for containers and CI. - Captcha and settings-panel callback payloads embed the owning user id and are rejected when pressed by anyone else.
- Every SQL query uses bound parameters; no SQL is built by string concatenation.
- Blacklist regular expressions are matched with a time limit so a hostile pattern cannot stall the event loop.
- The systemd unit ships a hardened profile; see
docs/deployment.md.