What Happened
Google Threat Intelligence Group and Mandiant warn of LLMjacking attacks, The Hacker News reports. Attackers steal LLM API keys or cloud identity tokens, then run models at the victim’s expense. Single incidents produced fraudulent charges above $100,000 in one day. Some organizations reached $1 million before they revoked access.
Initial access comes from phishing, leaked source code, exposed client-side code, stolen service account keys, and GitHub tokens. Threat actor UNC6993 used a fake portal to capture a user’s access token and bypass password and MFA controls. After access, attackers escalate privileges, scan cloud regions, harvest secrets, and create backdoor accounts and SSH keys. Resellers called LLM mixers sell the stolen access. One illicit service offers unlimited tokens for about $60 a month. Attackers also exploited CVE-2026-40217 in LiteLLM to escalate privileges.
Why This Matters for Canadian Organizations
Canadian startups, universities, hospitals, and banks are adding AI features fast. Many keep API keys in repositories, build pipelines, and shared notebooks. A stolen key turns into a cloud bill in hours, and finance teams learn about it at month end.
The risk goes past cost. The same cloud footprint holds customer data. An attacker with a foothold in your cloud account is a privacy incident under PIPEDA, and federally regulated firms must weigh OSFI B-13 expectations on third-party and technology risk. Budget teams funded through grants or public money carry added exposure when a bill spikes.
Read more in our Trends coverage.
What to Do
Search repositories and pipelines for hard-coded AI keys and rotate any you find. Use short-lived credentials and least-privilege service accounts. Enforce phishing-resistant MFA for cloud consoles. Set billing alerts and hard spending quotas on AI services. Audit for unknown service accounts, IAM keys, and SSH keys. Patch LiteLLM and other routing software, and keep it off the public internet. For more daily coverage, see our latest news.
