FABLE 5

### Preflight Checklist

- [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Amodel) for similar behavior reports
- [x] This report does NOT contain sensitive information (API keys, passwords, etc.)

### Type of Behavior Issue

Claude accessed files outside the working directory

### What You Asked Claude to Do

all. That is not a safety trade-off — that is a product that does not deliver what was sold.

What I want:

Narrow the [cyber] classifier so self-hosted personal development is not treated as attack tooling.
An opt-in verified-developer path for people building on their own hardware.
Stop billing safeguard-blocked "thinking" time against usage limits. Charging me for 15 minutes of computation that produces a refusal is indefensible.
If Fable 5 cannot serve ordinary Android development, say so plainly on the pricing page before people pay for it.

I am not asking you to loosen safety. I am asking you to stop treating gradlew assembleDebug as a cyberweapon.

### What Claude Actually Did

all. That is not a safety trade-off — that is a product that does not deliver what was sold.

What I want:

Narrow the [cyber] classifier so self-hosted personal development is not treated as attack tooling.
An opt-in verified-developer path for people building on their own hardware.
Stop billing safeguard-blocked "thinking" time against usage limits. Charging me for 15 minutes of computation that produces a refusal is indefensible.
If Fable 5 cannot serve ordinary Android development, say so plainly on the pricing page before people pay for it.

I am not asking you to loosen safety. I am asking you to stop treating gradlew assembleDebug as a cyberweapon.

### Expected Behavior

all. That is not a safety trade-off — that is a product that does not deliver what was sold.

What I want:

Narrow the [cyber] classifier so self-hosted personal development is not treated as attack tooling.
An opt-in verified-developer path for people building on their own hardware.
Stop billing safeguard-blocked "thinking" time against usage limits. Charging me for 15 minutes of computation that produces a refusal is indefensible.
If Fable 5 cannot serve ordinary Android development, say so plainly on the pricing page before people pay for it.

I am not asking you to loosen safety. I am asking you to stop treating gradlew assembleDebug as a cyberweapon.

### Files Affected

```shell

```

### Permission Mode

Accept Edits was ON (auto-accepting changes)

### Can You Reproduce This?

Yes, every time with the same prompt

### Steps to Reproduce

_No response_

### Claude Model

Other

### Relevant Conversation

```markdown
all. That is not a safety trade-off — that is a product that does not deliver what was sold.

What I want:

Narrow the [cyber] classifier so self-hosted personal development is not treated as attack tooling.
An opt-in verified-developer path for people building on their own hardware.
Stop billing safeguard-blocked "thinking" time against usage limits. Charging me for 15 minutes of computation that produces a refusal is indefensible.
If Fable 5 cannot serve ordinary Android development, say so plainly on the pricing page before people pay for it.

I am not asking you to loosen safety. I am asking you to stop treating gradlew assembleDebug as a cyberweapon.
```

### Impact

Critical - Data loss or corrupted project

### Claude Code Version

5

### Platform

Anthropic API

### Additional Context

_No response_

The report names no files or tests and provides no concrete reproduction steps beyond running `gradlew assembleDebug`. Start by clarifying which Claude Code behavior and safeguard decision occur in the supplied conversation, then separate the classifier, usage billing, verified-developer, and pricing requests. Done would require an agreed scope and acceptance criteria for the selected issue.

索引モデルが issue の本文から書いたものです。