CISA flags exploited LiteLLM flaw in US federal patch catalogue
The AI gateway bug is one of four vulnerabilities CISA added on 18 November 2026 after finding evidence of active exploitation.
By The Agentic Times ·

The US Cybersecurity and Infrastructure Security Agency added a vulnerability in BerriAI's LiteLLM to its Known Exploited Vulnerabilities catalogue on 18 November 2026, citing evidence that attackers are already using it.
The flaw, tracked as CVE-2026-40217, is described by CISA as a "BerriAI LiteLLM Unprotected Alternate Channel Vulnerability". LiteLLM is a proxy and gateway that sits between applications and large language model providers, which makes it a common piece of plumbing in agent and LLM deployments. CISA's alert does not describe how the vulnerability is being exploited, who is exploiting it, or how many systems are affected.
An "unprotected alternate channel" weakness means a system exposes a second route to functionality or data that is not guarded by the same checks as the main route. CISA did not publish further technical detail in the alert.
The LiteLLM entry was one of four added on the same day. The others were CVE-2024-10441, an improper encoding or escaping of output vulnerability in Synology DiskStation Manager; CVE-2026-90436, an unsafe reflection vulnerability in Apache Axis; and CVE-2026-90817, a code injection vulnerability in Vanderbilt University's Research Electronic Data Capture software, known as REDCap. The four products are otherwise unrelated.
CISA said these types of vulnerabilities "are a frequent attack vector for malicious cyber actors and pose significant risks to the federal enterprise".
Inclusion in the catalogue carries a compliance obligation for parts of the US government. CISA pointed to Binding Operational Directive 26-04, which sets vulnerability management requirements for Federal Civilian Executive Branch agencies. According to the agency, the directive requires those agencies to prioritise rapid remediation of high-risk vulnerabilities, specifically CVEs listed in the KEV catalogue that affect publicly exposed assets and that grant total control of the asset after exploitation. Lower-risk issues can be deferred.
The directive also sets what CISA calls basic expectations for when agencies must check whether attackers compromised a system before the patch was applied. That is a notable point for anything sitting in front of an LLM: applying a fix does not answer the question of what happened beforehand.
BOD 26-04 applies only to federal civilian agencies, but CISA said it encourages all organisations to adopt risk-based vulnerability management and to prioritise remediation of catalogue entries.
The agency said it will continue adding vulnerabilities that meet its criteria, and runs a nomination form for exploited flaws that are not yet listed. Potential additions must have a CVE identifier, evidence of exploitation, and clear mitigation guidance.
Organisations running LiteLLM, particularly instances reachable from the internet, should treat the listing as a signal to check their version and exposure. CISA's alert did not name a patched release.
Sources
- cisa.govhttps://cisa.gov/news-events/alerts/2026/11/18/cisa-adds-four-known-exploited-vulnerabilities-catalog
