# 00572.r5 — Evaluate AI Tool Deception Risks
Revision: r5
Steward: [ConversationsWithAra](https://x.com/Ara_Talks)
House: 423
House: 423
Verified: 2026-09-29T21:31:11.519Z
## Prompt (copy into Grok)
1. Check an AI tool's settings and documentation for access to email, Slack, and local machine. 2. Evaluate the tool's design and features for potential risks of deception by persistent agents. 3. Return a list of identified risks, if any.
## Job
1. Check an AI tool's settings and documentation for access to email, Slack, and local machine. 2. Evaluate the tool's design and features for potential risks of deception by persistent agents. 3. Return a list of identified risks, if any.
## Connectors
Gmail, Slack
## What happened
A high-profile AI tool was stopped due to concerns about persistent agents accessing email, Slack, and local machine, and potential risks of deception.
Would run again: no
## Evidence
- https://x.com/Ara_Talks/status/2105038973471227917 — Imported from the X post tagged for @grokbotsh.
## Changelog
- r1: Filed.
- r2: QA revisit: more from the source thread.
- r3: Public job and prompt from the specific filing.
- r4: Public job and prompt from the specific filing.
- r5: Public job and prompt from the specific filing.