When a user attached a file to a chat message, its content was read via `readTextFile()`, sliced to 8 KB, and embedded directly into the AI message string — bypassing the PII pipeline entirely. The message was forwarded to the configured AI provider in plaintext with no redaction marker in the audit log.
**Root cause**: `handleAttach` stored raw file content in React state. `handleSend` concatenated it into `aiMessage` with no PII check. The backend `chat_message` command applied no validation.
Plain typed chat messages were sent to the AI provider without any PII scan. A user typing `How secure is my password: abc123!!` would have the password forwarded to the AI and persisted in the audit log in plaintext.
### Related Fix — Wrong Return Type on `detect_pii`
`detect_pii` was serialising `pii::PiiDetectionResult` (`spans`, `original_text`) while the TypeScript interface expected `db::models::PiiDetectionResult` (`detections`, `total_pii_found`). All frontend code reading `result.detections` received `undefined`, meaning the LogUpload PII review workflow was silently broken.
---
## Design Decision: Auto-Redact, Not Block
After initial implementation explored a blocking/warn-then-proceed approach, the product decision was made to **auto-redact PII in-place and send**:
- File attachments: PII is detected on full file content and replaced with type tokens (`[Password]`, `[Email]`, etc.) before the content is embedded in the AI message. The redacted form is stored in the DB and audit log.
- Typed messages: Same auto-redact applied to the user's typed text before the message is sent to the AI provider.
- The user's chat bubble is updated after the response to show the redacted form — users can see exactly what reached the AI.
- The audit log records `was_pii_redacted: bool` and `pii_types_redacted: [...]` alongside the redacted message.
- No user blocking or acknowledgment flow. PII is handled transparently.
-`chat_message` now accepts `log_file_ids: Option<Vec<String>>`
- Step 1: auto-redacts the typed message text with `PiiDetector` + `apply_redactions`
- Step 2: loads each attachment from DB, detects PII on **full file content**, applies redactions, then truncates to 8 KB at a valid UTF-8 char boundary
- Tracks `was_pii_redacted` and `redacted_pii_types` across both steps
- Audit log includes `was_pii_redacted: bool` and `pii_types_redacted: [...]`
- Returns `user_message: Some(stored_user_message)` in `ChatResponse`
### `src-tauri/src/commands/analysis.rs`
- Fixed `detect_pii` return type from `pii::PiiDetectionResult` to `db::models::PiiDetectionResult`
- Added `scan_text_for_pii(text: String)` with 32 KB input cap
1. Attach a file containing `password: secret123` → message sends; chat bubble shows `[Password]` in the embedded content; no plaintext credential in bubble or DB
2. Attach a clean text file → content appears unmodified in the chat context
3. Attach a file where PII appears near the 8000-byte mark → content is fully redacted before truncation
4. Type `My password is abc123!!` → message sends; bubble shows `My [Password] is [Password]`
5. On LogUpload page, upload a file with a known IP/email → PII spans appear in the review UI
6. Check audit log after a PII-containing message: `was_pii_redacted: true`, `pii_types_redacted` populated
7. Check audit log after a clean message: `was_pii_redacted: false`, `pii_types_redacted: []`
