CVE-2026-59822: BerriAI LiteLLM Improper Authentication — CISA KEV Addition & LLM Proxy Compromise
Technical teardown of CVE-2026-59822 in BerriAI LiteLLM: authentication bypass in enterprise AI proxies, LLM key exfiltration, and CISA KEV implications.
CISA has added an improper authentication vulnerability in BerriAI LiteLLM (CVE-2026-59822) to its Known Exploited Vulnerabilities catalog, marking a critical milestone in nation-state and cybercrime targeting of enterprise AI proxy and LLM gateway infrastructure.
CVE-2026-59822 (CWE-287: Improper Authentication)
Severity: Critical (CVSS 9.8)
Status: 🚨 Added to CISA KEV Catalog / Active AI Gateway Exploitation
Affected Systems: BerriAI LiteLLM proxy deployments prior to version 1.44.15
Threat Context: Weaponizing the AI Gateway
LiteLLM serves as an orchestration proxy forwarding developer prompts to upstream LLM providers (OpenAI, Anthropic, AWS Bedrock). An authentication bypass in the LiteLLM admin dashboard API allows unauthenticated remote actors to query internal model endpoints, exfiltrate API keys, and poison conversational context memory.