fix: type error & better error handling

[Harshit28j]

Relevant issues

Presidio API returned a plain string (e.g. an error message) instead of a list of dicts. That often occurred with web search or hosted models. The code treated analyze_results as a list and iterated over it; for a string, it iterated over characters and tried PresidioAnalyzeResponseItem(**"e"), which triggered the error.

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have Added testing in the tests/litellm/ directory, Adding at least 1 test is a hard requirement - see details
• My PR passes all unit tests on make test-unit
• My PR's scope is as isolated as possible, it only solves 1 specific problem

CI (LiteLLM team)

CI status guideline:

• 50-55 passing tests: main is stable with minor issues.
• 45-49 passing tests: acceptable but needs attention
• <= 40 passing tests: unstable; be careful with your merges and assess the risk.

• Branch creation CI run
  Link:
• CI run for the last commit
  Link:
• Merge / cherry-pick CI run
  Links:

Type

🐛 Bug Fix
✅ Test

Changes

Code changes

• Unsupported types: Added a check so that if analyze_results is not a list (e.g. string, None), the code logs a warning and returns [].
• List iteration: Before PresidioAnalyzeResponseItem(**item), added a check that item is a dict. Non-dict items are skipped with a warning.
• Exception handling: The try/except around PresidioAnalyzeResponseItem(**item) now catches all parsing errors (not just TypeError), logs them, and continues

ref docs on this:
-https://microsoft.github.io/presidio/api/analyzer_python/
-https://microsoft.github.io/presidio/api-docs/api-docs.html
-https://microsoft.github.io/presidio/api-docs/api-docs.html

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Feb 11, 2026 2:02am

[greptile-apps]

Greptile Overview

Greptile Summary

This PR hardens the Presidio guardrail hook’s /analyze response handling so it no longer crashes when Presidio returns malformed JSON shapes (e.g., a plain string error message, None, or non-dict items inside a list). It adds type checks to treat non-list responses as empty detections and skips invalid list items while continuing to parse valid ones. New unit tests cover string responses and mixed-type lists by mocking the analyzer API response.

Confidence Score: 3/5

• This PR is likely safe but needs a guardrail-failure decision for malformed Presidio responses and slightly more robust tests.
• Core change is small and targeted (type checks + skip invalid items) and addresses a real crash scenario. However, defaulting to [] for unexpected analyzer response types can silently disable blocking behavior if Presidio is misconfigured or intermittently failing. Tests also patch aiohttp directly, which may not exercise the actual session abstraction used by the hook in all configurations.
• litellm/proxy/guardrails/guardrail_hooks/presidio.py; tests/test_litellm/proxy/guardrails/guardrail_hooks/test_presidio.py

Important Files Changed

Filename	Overview
litellm/proxy/guardrails/guardrail_hooks/presidio.py	Adds defensive handling for Presidio analyzer responses that are non-lists and skips non-dict items; main concern is that returning [] on unexpected types can silently disable blocking guardrails if Presidio misbehaves.
tests/test_litellm/proxy/guardrails/guardrail_hooks/test_presidio.py	Adds tests for string analyzer responses and mixed-type lists; tests may be brittle because they patch aiohttp.ClientSession rather than the code’s session abstraction (_get_session_iterator), so they may not exercise the intended path if session reuse/close logic changes.

Sequence Diagram

sequenceDiagram
  participant Hook as PresidioPIIMasking
  participant Sess as aiohttp ClientSession
  participant Analyzer as Presidio Analyzer API

  Hook->>Sess: POST {analyzer_base}/analyze (analyze_payload)
  Sess->>Analyzer: HTTP request
  Analyzer-->>Sess: JSON body (list|dict|string|null)
  Sess-->>Hook: analyze_results = response.json()

  alt analyze_results is dict
    Hook->>Hook: if 'error' in dict => return []
    Hook->>Hook: else try PresidioAnalyzeResponseItem(**dict)
  else analyze_results is not list
    Hook->>Hook: warn + return []
  else analyze_results is list
    loop each item
      alt item is dict
        Hook->>Hook: parse PresidioAnalyzeResponseItem(**item)
      else item not dict
        Hook->>Hook: warn + skip
      end
    end
    Hook-->>Hook: return final_results
  end

Loading

[greptile-apps]

_{2 files reviewed, 2 comments}

_{Edit Code Review Agent Settings | Greptile}

[krrishdholakia]

@greptile please review this

[shivamrawat1]

@greptile review this

[greptile-apps]

Greptile Overview

Greptile Summary

Fixes a TypeError crash when the Presidio analyzer API returns a non-list response (e.g., a plain string error message). Adds type checks for analyze_results and individual list items, and introduces fail-closed semantics (GuardrailRaisedException) when PII protection is configured, preventing silent bypass of PII detection on malformed responses.

• Added _fail_on_invalid_response helper that raises GuardrailRaisedException when pii_entities_config, output_parse_pii, or apply_to_output is set, and falls back to empty list otherwise
• Added type guard for non-list/non-dict analyze_results (handles string, None, etc.)
• Added per-item isinstance(item, dict) check when iterating list results
• Broadened except TypeError to except Exception for parsing failures
• Added 5 new mock-only tests covering string responses, error dicts, fail-closed behavior, and mixed-type list items
• _has_block_action() method was added but is unused — consider removing it

Confidence Score: 4/5

• This PR is safe to merge — it adds defensive error handling to an existing code path with appropriate fail-closed behavior.
• The changes are well-scoped, correctly implement fail-closed semantics when PII protection is configured, and include comprehensive test coverage. Minor issue: one unused method (_has_block_action). No risk of regression to existing behavior.
• No files require special attention beyond the minor dead code in litellm/proxy/guardrails/guardrail_hooks/presidio.py.

Important Files Changed

Filename	Overview
litellm/proxy/guardrails/guardrail_hooks/presidio.py	Adds robust error handling for invalid Presidio API responses (string, None, malformed dict/list). Implements fail-closed semantics via GuardrailRaisedException when PII protection is configured. Contains one unused method _has_block_action().
tests/test_litellm/proxy/guardrails/guardrail_hooks/test_presidio.py	Adds 5 new test cases covering string responses, error dicts, fail-closed with BLOCK/MASK configs, and mixed list items. Uses a well-structured _make_mock_session_iterator helper that patches _get_session_iterator directly. All tests are mock-only.

[greptile-apps]

_{2 files reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_has_block_action() is defined but never called anywhere in the codebase. The fail-closed logic in _fail_on_invalid_response uses bool(self.pii_entities_config) instead. Remove this dead code or use it in the should_fail_closed check if the intent was to only fail-closed on BLOCK actions.

[ishaan-jaff]

lgtm