claude-code - 💡(How to fix) Fix AUP false positives on government accountability / open data analysis [1 comments, 2 participants]

Official PRs (…)
ON THIS PAGE

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

Helpful · Quick feedback

Loading…
GitHub stats
anthropics/claude-code#56319Fetched 2026-05-06 06:31:18
View on GitHub
Comments
1
Participants
2
Timeline
1
Reactions
0
Timeline (top)
commented ×1

Fix Action

Workaround

Phrasing queries in more technical/database terms ("query iv3_municipal_finance for taakveld 5.4") instead of natural language ("where is the culture money going") seems to reduce false positives, but this shouldn't be necessary.

RAW_BUFFERClick to expand / collapse

Problem

When using Claude Code to work on a Dutch government spending transparency project (dutch-gov-accountability), I repeatedly hit AUP (Acceptable Use Policy) violations for completely legitimate queries.

The project collects publicly available open data from:

  • CBS (Centraal Bureau voor de Statistiek) — Dutch national statistics
  • Rijksfinancien — Central government budget data
  • Iv3 (dataderden.cbs.nl) — Municipal finance data for all 342 Dutch municipalities

All data sources are official government APIs designed for public transparency and accountability.

What triggers it

Even simple investigative queries like "where is the culture money going" (i.e., querying municipal spending on taakveld 5.4 Musea) trigger AUP violations. The intent is analyzing public expenditure data that the Dutch government publishes specifically for civic oversight.

Expected behavior

Analyzing publicly available government spending data for accountability purposes should not trigger AUP flags. This is exactly the kind of civic transparency work these datasets are published for.

Workaround

Phrasing queries in more technical/database terms ("query iv3_municipal_finance for taakveld 5.4") instead of natural language ("where is the culture money going") seems to reduce false positives, but this shouldn't be necessary.

extent analysis

TL;DR

Rephrasing queries in more technical terms may help reduce Acceptable Use Policy (AUP) violations when analyzing public government spending data.

Guidance

  • Verify that the issue persists across different types of queries and data sources (CBS, Rijksfinancien, Iv3) to rule out source-specific problems.
  • Test the workaround of using technical terms in queries to see if it consistently reduces AUP flags.
  • Consider reaching out to the API providers or Claude Code support to report the issue and gather more information on their AUP policies and potential false positive reduction strategies.
  • Document the queries that trigger AUP violations and those that do not, to identify any patterns or common characteristics.

Notes

The root cause of the AUP violations is unclear, and it is uncertain whether the workaround will be effective in all cases. Further investigation and communication with the API providers or Claude Code support may be necessary to fully resolve the issue.

Recommendation

Apply the workaround of rephrasing queries in more technical terms, as it has shown some promise in reducing AUP flags, and continue to monitor the issue to see if further adjustments are needed.

Vote matrix · Quick signals

Works
Did the solution work? Tap to confirm.
Easy Fix
Was it a quick fix?
Time Saver
Did it save you time?
Blocking
Was it severely blocking?
Common Issue
Are others likely hitting this too?
Flaky / Intermittent
Is it intermittent?
Verified / Reproducible
Can you reproduce it reliably?
Loading…

FAQ

Expected behavior

Analyzing publicly available government spending data for accountability purposes should not trigger AUP flags. This is exactly the kind of civic transparency work these datasets are published for.

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

Back to top recommendations

TRENDING