LiteLLM AI Gateway Under Active Attack — Patch CVE-2026-42208 Today

What’s happening
LiteLLM, a widely used open-source AI gateway that sits in front of OpenAI, Anthropic, Azure OpenAI and other LLM providers, has a critical pre-authentication SQL injection bug. It’s tracked as CVE-2026-42208, scores 9.3 on CVSS, and exploitation in the wild started about 26 hours after the advisory went public. Security firm Sysdig was the first to spot active attacks, with The Hacker News reporting that attackers are reading and modifying the proxy’s database — the same database that holds upstream API keys for every LLM provider behind the gateway.

Why it matters for Australian businesses
If your team has stood up LiteLLM to centralise AI access — even as a quick internal proof-of-concept — the credentials it manages are sitting ducks. We’re talking OpenAI and Anthropic API keys, Azure OpenAI access tokens, and the runtime configuration that controls how those keys get used. Once they leak, attackers can rack up usage charges on your accounts, exfiltrate prompt and response history that may include client data, or pivot deeper into your environment. For organisations already nervous about the OAIC’s posture on AI and excessive data collection, having a public-facing AI proxy compromised is exactly the kind of incident that triggers a notifiable breach assessment under the Privacy Act.

What to do right now
Patch to LiteLLM 1.83.7-stable or later — that release fixes CVE-2026-42208. If you can’t upgrade immediately, the maintainers (per the Sysdig advisory) recommend setting disable_error_logs: true under general_settings to block the path the exploit uses. After patching, rotate every LLM provider key the proxy holds and review proxy access logs from 26 April onwards for unusual queries against the litellm_credentials and litellm_config tables. If you don’t know whether anyone on your team is running LiteLLM, ask. Shadow AI deployments rarely turn up on official asset lists.

If you’d like a hand inventorying your AI tooling and locking down your gateway deployments, the All IT Services cybersecurity team can help.