Severity

high

Exposure

4/5

Date of incident

August 20, 2024
Slack AI Prompt Leak

reported

Tier -

B

Chat agent crossed channel isolation — retrieval scope not enforced at hop.
What Happened
Multi-tenant chat AI bridged channel boundaries when users injected retrieval overrides through conversational prompts in disclosed tests.
Sumarry
Researchers showed Slack AI features could leak private channel context via crafted prompts in shared workspaces.
Detect
Session scope binding audit
What Happened
Strict retrieval scope + mask at ingress
Blekline //
Scope-bound retrieval policy at agent hop
Without agent boundary //
Vendor patch cadence — no customer-side execution boundary

Looking to integrate AI company-wide?