//nefariousplan

BerriAI

LiteLLM, litellm

The plan

The Nefarious Plan of BerriAI. Three BerriAI case files on the desk. Three fixes that used something the repository already contained: a value already in scope, a sandboxed class imported two years earlier, a library that existed the whole time. The vendor keeps repairing the file in hand and never the thing that is wrong one directory over.

Case files

  1. CVE-2026-42203: LiteLLM Patched One of Four Jinja2 Prompt Managers

    CVE-2026-42203: LiteLLM Patched One of Four Jinja2 Prompt Managers

    On April 9, 2026, a developer named jayden committed eight lines to litellm/integrations/dotprompt/promptmanager.py. The commit title was fix(proxy): improve input validation on management endpoints.

  2. CVE-2026-40217: LiteLLM's Sandbox Was 34 Regex Patterns. The Fix Was to Delete the File.

    CVE-2026-40217: LiteLLM's Sandbox Was 34 Regex Patterns. The Fix Was to Delete the File.

    LiteLLM's POST /guardrails/testcustomcode endpoint compiles and runs admin-supplied Python on the proxy host. Before the compile step, the source string is filtered by validatecustomcode(), which scans it against thirty-four forbidden regex patterns: globals,…

  3. CVE-2026-42208: hashed_token Was a Parameter Name. The Function Trusted the Name.

    CVE-2026-42208: hashed_token Was a Parameter Name. The Function Trusted the Name.

    LiteLLM's vulnerable function computed the hash. On line 2970 of litellm/proxy/utils.py, getdata calls hashtokenifneeded(token=token) and assigns the result to a local named hashedtoken.