After an agent leaked salary data, Reddit distilled a 5-step pre-launch access checklist
Wild-Lawyer8511 · reddit · 2026-10-08
A company bot meant to read only a wiki indexed everything—including an HR folder—after its setup person accepted Drive access, and answered questions with an unannounced reorg plan and salary bands; another summarizing agent inherited delete access on a shared drive and could send email as its creator. Though the original post was removed, its 70+ replies distilled into a pre-launch checklist: give the agent its own identity (and check access against the asker at query time); pull real grants from the admin console, not memory; keep reading and doing apart with separate narrow credentials; ask it questions it should refuse before launch; and test cached answers after revoking access. No new tools needed—just treat it like a smart but drunk intern.
More from coding & agent
- OpenEnv Arena guide: let your agent check metrics and climb the RL environment leaderboard — ben_burtenshaw · 2026-10-08
- 192 local runs show even good models shouldn't authorize their own tool calls — Slight_Analysis_5414 · 2026-10-08
- Jev 0.4.0 brings semantic ranking and routing to PowerShell pipelines — dfinke · 2026-10-08
- Five AI agents handle a mid-project requirement change in human-agent platform Teamily — cneuralnetwork · 2026-10-08
- Watching my AI agent do my job while I just say 'looks good' and 'continue' — tekbog · 2026-10-08
- Getting an LLM judge from garbage to 27/30 human agreement: an evals workshop writeup — tejaskumarlol · 2026-10-08