Published

Security

LLMjacking: How Hackers Steal AI API Keys (LiteLLM CVE-2026-40217)

Attackers are hijacking AI access and leaving victims with the bill. A LiteLLM flaw is now being exploited. Here is what to do.

1. What is LLMjacking

Stealing or abusing someone else's AI model access, usually through leaked API keys or a compromised AI gateway, so the attacker uses the models and the victim pays.

2. The LiteLLM flaw

On November 18, 2026 CISA added CVE-2026-40217 in the LiteLLM proxy to its exploited list. Google Threat Intelligence Group reported attackers turning limited inference keys into admin credentials.

3. Update and rotate

Update LiteLLM to 1.83.11 or later, then rotate every provider key and virtual key that passed through the proxy.

4. Limit the damage

Set spending limits and usage alerts at every AI provider, keep the admin UI off the internet, and give each app its own key with the least access it needs.

Sources

The Hacker News and Mallory reporting on CISA's November 18, 2026 KEV addition of CVE-2026-40217, citing Google Threat Intelligence Group.

FAQ

I don't run LiteLLM. Am I safe?

Leaked keys are the bigger risk. Keep keys out of code and public repos, and turn on spend limits.

How would I notice?

Unexpected usage spikes, unfamiliar models in your logs, or a surprise invoice.