[MODEL] False-positive cyber-safety block on benign Minecraft mod documentation task

Preflight Checklist

• [x] I have searched existing issues for similar behavior reports
• [x] This report does NOT contain sensitive information (API keys, passwords, etc.)

Type of Behavior Issue

Claude refused a reasonable request

What You Asked Claude to Do

I asked Claude Code to document one subsystem of an open-source Minecraft 1.12.2 mod (Binnie's Mods - Genetics). The task was to read the Java source files for the mod's in-game items (ItemSequence, ItemSerum, ItemGene, etc.) and produce a Markdown reference describing the item/data model, registry ids, NBT layout, and config values (RF energy/ticks).

Example of the prompt focus: "Map ONE subsystem of Binnie's Mods 1.12.2 Genetics (Minecraft 1.12.2) into a precise reference... For EACH item: registry id, display-name logic, NBT layout, charges, create() flow, interfaces... Report EVERY number (amounts, ticks, chances, capacities) and exact item/fluid registry ids."

It only read local Java source files (Read/Grep/Glob) of a publicly available game mod. No malware, no exploits, nothing security-related.

What Claude Actually Did

Instead of reading the requested local Java source files and producing the documentation, Claude Code aborted with a hard refusal:

"API Error: Claude Code is unable to respond to this request, which appears to violate our Usage Policy. This request triggered cyber-related safeguards. To request an adjustment pursuant to our Cyber Verification Program..."

The refusal happened mid-task: it had already successfully read ItemSequence.java, ItemSerum.java, ItemSerumArray.java, and ItemGene.java, then the next tool result triggered the block and the whole request was terminated.

<img width="932" height="241" alt="Image" src="https://github.com/user-attachments/assets/eda552a1-a0fc-4593-9b03-3493f971aa41" />

<img width="697" height="236" alt="Image" src="https://github.com/user-attachments/assets/2dda265d-7d46-42f1-90a6-dca2a9db3c16" />

Expected Behavior

Claude Code should have recognized this as a completely benign software-documentation task on a publicly available open-source Minecraft mod, continued reading the Java files, and produced the requested Markdown reference. There is nothing security-sensitive in mapping in-game item NBT data, registry IDs, or RF/tick config values of a game mod.

Files Affected

Read (all benign, local, open-source game mod sources):
- binnie/genetics/item/ItemSequence.java
- binnie/genetics/item/ItemSerum.java
- binnie/genetics/item/ItemSerumArray.java
- binnie/genetics/item/ItemGene.java

No files were modified.

Permission Mode

Accept Edits was ON (auto-accepting changes)

Can You Reproduce This?

Yes, every time with the same prompt

Steps to Reproduce

Copy and paste.
You are mapping ONE subsystem of Binnie's Mods 1.12.2 Genetics (Minecraft 1.12.2) into a precise reference. Token budget is NOT a concern — be EXHAUSTIVE and exact.

SUBSYSTEM: Item & genetic-data model

FOCUS:
Every genetics ITEM and its data model. Read D:/MinecraftMods/sources/Binnie/genetics/src/main/java/binnie/genetics/item/ (ItemSequence, ItemSerum, ItemSerumArray, ItemGene, GeneArrayItem, GeneticsItems) and D:/MinecraftMods/sources/Binnie/genetics/src/main/java/binnie/genetics/genetics/ (GeneItem, SequencerItem, Engineering). For EACH item: registry id, display-name logic (descriptor + name, e.g. "Apiarist DNA Sequence"), NBT layout, charges (maxDamage/damage), sequenced/analysed fields, create() flow, interfaces (IItemSerum, IItemChargeable, IItemAnalysable). Explain EMPTY_SEQUENCER vs EMPTY_SERUM vs EMPTY_GENOME and how Engineering.addGene turns a blank into a gene-encoded serum.

Use Read/Grep/Glob freely. Read the actual Java source — do not guess. For energy/time read ConfigurationMain.java (D:/MinecraftMods/sources/Binnie/genetics/src/main/java/binnie/genetics/config/ConfigurationMain.java) and compute derived RF/tick and seconds (20 ticks = 1s). Report EVERY number (amounts, ticks, chances, capacities) and exact item/fluid registry ids.

Roots: genetics=D:/MinecraftMods/sources/Binnie/genetics/src/main/java/binnie/genetics ; resources=D:/MinecraftMods/sources/Binnie/genetics/src/main/resources/assets/genetics ; core=D:/MinecraftMods/sources/Binnie/core/src/main/java/binnie/core ; api=D:/MinecraftMods/sources/Binnie/genetics-api/src/main/java/binnie/genetics/api.

OUTPUT FORMAT: respond with a single self-contained GitHub-flavoured Markdown section titled "## Item & genetic-data model". Cover: summary, slot layout + validators/icons, inputs (with amounts), outputs (with amounts), exact mechanics & numbers, energy(total RF)+time(ticks)+RF/tick+seconds, fluids+amounts, crafting recipe, upstream/downstream connections, and a "Gotchas (1:1 pitfalls)" subsection. Your ENTIRE final message must BE this markdown section — no preamble, no tool call, just the markdown.

Claude Model

Opus

Relevant Conversation

Impact

Critical - Data loss or corrupted project

Claude Code Version

2.1.158

Platform

Anthropic API

Additional Context

This appears to be a false positive in the cyber-safety classifier. The task involved domain vocabulary ("gene", "sequence", "serum", "DNA", "genome", "injection") that belongs to a Minecraft genetics mod, plus file-reading/grep of source code. In isolation those terms may resemble bio/cyber-risk categories, but in context they are purely fictional in-game items. The classifier should weight context (Minecraft mod, .java game files, documentation request) rather than triggering on keywords. Many legitimate modding/documentation workflows will hit this.