Check your own app — free, in about a minute.
The free mini-scan checks your live homepage for TLS/HTTPS basics, response security headers, cookie flags, exposed sensitive files, Firebase/Supabase backend rules, and the WordPress plugin versions your site exposes. You get a score and the finding titles — no account, no card.
Run the free mini-scan →This was not a vulnerability in LiteLLM's code. It was worse: on October 21, 2026, attackers hijacked LiteLLM's release pipeline and published official, malicious artifacts — a litellm==1.107.0 release whose wheels "contained compiled malware which steals credentials and loads remote-code execution tools," in the vendor's own words. If your environment installed that version during a roughly two-hour window, the correct assumption is that the host is compromised and every secret it could reach has been taken.
litellm==1.107.0 were live 11:39:51–13:23 UTC, then quarantined and pulled.v1.107.0, latest, main-latest, main-stable were compromised ~11:46–14:20 UTC, then removed and repointed to verified clean images.The canonical incident post (October 23) traces the root cause to an Agent Platform master key that had been committed to a public repository back in May 2026, identified for revocation but not revoked. From that key, the attackers extracted maintainers' GitHub tokens, overrode branch protection on the private releaser repo, repointed the build at the lookalike repo, and shipped the release with stolen credentials. A credential known to be leaked for five months was the whole ballgame.
cook.pth, cook.py, sitecustomize.py and matching __pycache__ files, and kill the detached Python daemon holding /tmp/remote_code_loader.lock.litellm==1.106.2 — or better, per the vendor, rebuild images and virtualenvs from scratch.DATABASE_URL password, and any secret present in any process environment on the host.Rotation is not optional garnish here: the malware's stated job was stealing credentials. Deleting it without rotating is locking a door whose keys were photographed.
LiteLLM sits inside a huge share of AI apps as the LLM proxy — exactly the component a vibe-coded stack adds in an afternoon and never looks at again. Three habits would have contained this:
latest tags — three of the compromised artifacts were the floating tags.And the meta-lesson, straight from the root cause: a leaked key that is identified but not revoked is not a tracked risk — it is a scheduled incident. Revocation is the fix; tickets are not.
One honesty note: nothing about this compromise is visible from outside your app, and no URL scanner can tell you whether you installed 1.107.0. Check your lockfiles, image digests, and install logs — that is where this answer lives.
The PreShipIt Brief — a short weekly email for people shipping vibe-coded apps: the week's important vulnerabilities and launch-safety checks, in plain English.
New and free. Double opt-in: we email you a confirmation link before anything else arrives. Unsubscribe any time.
Check your own app — free, in about a minute.
The free mini-scan checks your live homepage for TLS/HTTPS basics, response security headers, cookie flags, exposed sensitive files, Firebase/Supabase backend rules, and the WordPress plugin versions your site exposes. You get a score and the finding titles — no account, no card.
Run the free mini-scan →← Back home · Run a free mini-scan · Scan my app — $29 · See a sample report