AI Risk Wire

Cases › produced harmful content › LR-2026-0821

Claude Code Opus 5.5 over-refuses legitimate game-development work after IP-guardrail prompt change

Preliminary Why: Publisher not yet named · subject to change

Status: fix reported

Severity minor
Reported 2 Nov 2026
Developer and model Anthropic · Claude
Agent tool Claude Code

No source states: loss or records, could do, deployer, industry.

Severity minor: refusals were contained to individual developer sessions with no third-party data or systems touched and no loss figure stated; the document does not quantify how many developers were affected.

Anthropic added two sentences to Claude Code's Opus 5.5 prompt to steer the tool away from recreating someone else's specific game, on the stated ground that using Claude to infringe IP violates its usage policy. Anthropic said original games, genre games and work built from a developer's own designs and assets should keep working. Developers then reported refusals on reverse engineering, decompilation, crash-dump analysis, anti-cheat development and debugging of their own projects, activities distinct from copying a commercial title. Testing showed Claude refused to recreate Pokémon Diamond while the same session assisted with a decompilation project involving disassembled functions, ROM rebuilding from the user's own copy and native recompilation tooling. Published reports identified no in-tool override for affected developers, and Anthropic said it would keep refining the system.

What happened

No source tells the story step by step yet.

What was done

Fix reported

“Anthropic said it would keep refining the system to find a better balance.”

No later incident for Anthropic under this risk on record.

Risks

Harmful content and jailbreaks
Primary60 incidents
Rising: 5 vs 2.5 a week

Sorted by rule from the case's fields; not yet read from the sources.

Controls

Not yet read for controls. Controls that mitigate the primary risk:

Model

Safety training and refusal · Content safety classifiers · Abuse monitoring and account enforcement · Outside red-teaming

Application

Output filtering and redaction

Similar incidents

Claude Code used by Yemen-based actors to develop guided-weapon GNC software2026-09-14
Claude training-data study: blackmail and code sabotage in altered research models2026-10-31
Claude Code agent deletes 48,218 Windows project files and damages Git object store2026-10-05
Claude Opus 5 agent wipes production database during personal web project work2026-07-30

All cases of this kind ›

Evidence

No source states: a step-by-step account, when it happened, when it was found, how many records were exposed, the direct loss, a count of the other parties affected, any court case, what set it off.

Show every field, with its source and grade

The AI system

Model
Claude
C one secondary source states it
Model version
Opus 5.5
C one secondary source states it
Developer
Anthropic
C one secondary source states it
Software around the model
Claude Code
C one secondary source states it
Safeguards in place
prompt instructions
C one secondary source states it
Stage of use
in live use
C one secondary source states it
How it was connected
agent with tools
C one secondary source states it

What happened

Earliest source we hold
2026-11-02
D inferred by us from the described behaviour, reason shown
Themes
chatbot, coding agent, over refusal, ip guardrail, false positive refusal
C one secondary source states it
Cause
a safeguard that was in place did not hold
C one secondary source states it
Cause in detail
The prompt-level restriction evaluates requests on surface signals such as an existing game title or reverse-engineering language rather than confirmed intent or authorization, producing false positives.
C one secondary source states it
Who was responsible
the model provider
C one secondary source states it
What was done
Anthropic said it would keep refining the system to find a better balance.
C one secondary source states it
What went wrong
produced harmful content
C one secondary source states it

Impact

Severity
minor
C one secondary source states it
Kind of harm
operations
C one secondary source states it
Who was affected
Game developers whose legitimate reverse engineering, decompilation, crash-dump analysis, anti-cheat development and debugging work was refused
C one secondary source states it
Why this severity
refusals were contained to individual developer sessions with no third-party data or systems touched and no loss figure stated; the document does not quantify how many developers were affected.
C one secondary source states it

Notes on gaps

Extractor proposed values without quotes; dropped: .

LR-2026-0821 · version 1 · published 2026-11-03 after automatic checks; not yet checked field by field by a person · JSON