Authenticated guardrail bypass via call-type mismatch
Published Jun 21, 2026 · Updated Jun 21, 2026
Improper authorization in LiteLLM 1.82.0 through 1.82.5 allows remote authenticated users to bypass banned-keyword filtering. The async_pre_call_hook compares the proxy's acompletion call type only with completion, so HTTP chat requests skip test_violation entirely. Exposure requires the enterprise BannedKeywords callback on the proxy; crafted prompts then reach the configured LLM backend despite the blocklist.
Summary
What happened
Improper authorization in LiteLLM 1.82.0 through 1.82.5 allows remote authenticated users to bypass banned-keyword filtering. The async_pre_call_hook compares the proxy's acompletion call type only with completion, so HTTP chat requests skip test_violation entirely. Exposure requires the enterprise BannedKeywords callback on the proxy; crafted prompts then reach the configured LLM backend despite the blocklist.
The record
- CVE
- CVE-2026-12797
- Published
- Jun 21, 2026
- Updated
- Jun 21, 2026
- Vendor
- Berri AI
- Product
- LiteLLM
- Classifications
- CWE-863, CWE-285
- Attack vector
- network
- Privileges
- authenticated
Timeline
How it unfolded
- Jun 21, 2026CVE publishedPublication date reported by the CVE source.
- Jun 21, 2026Record updatedLatest update available in the CVE record.
Exploitability
Present is not the same as exploitable
Compare your product and version with the public record. A matching version still requires validation against your environment.
Is a vulnerable build present?
Compare these published version ranges with your installed build and any vendor patches.
- Affected versionversion=1.82.0
- Affected versionversion=1.82.1
- Affected versionversion=1.82.2
- Affected versionversion=1.82.3
- Affected versionversion=1.82.4
- Affected versionversion=1.82.5
What conditions does exploitation require?
What is affected?
Attacks
What attackers are doing with it
Daily unique IPs observed by Shadowserver honeypots for known exploited vulnerabilities (KEVs). Missing observations do not establish an absence of attacks.
Public exploit references
- Security Feature Bypass in BannedKeywords and AzureContentSafety Guardrails via call_type Mismatch on Async Endpointsproof of concept · demonstrated
Labels summarize the accepted research assessment. They do not indicate a test against your environment.
Technologies
Your stack
See the directory against your own environment.
Your stack
Check the software in your environment
Book a demo to see how Hinoki identifies affected software and validates exploitability in your environment.
Book a demo