vulns.co
/
GKData.io MCP and API Buy me a coffee

Google · AI integration boundaries

Bard Workspace integration weakened output-data boundaries

Bard's Workspace integration could disclose connected email content through generated output. Researchers demonstrated this with a controlled account and report a USD 20,000 base award for the finding. That award covers one finding within a broader article.

Read the original writeupUSD 20,000 reported award · Reviewed 2026-10-02

Root cause

Generated content and browser output restrictions did not maintain the intended boundary around connected Workspace data.

Demonstrated impact

The researchers demonstrated disclosure of email content from a controlled account.

What to check for

  • Independently constrain external destinations for sensitive model output.
  • Review connector data access and rendering as a single end-to-end trust boundary.

Award and evidence

USD 20,000Bug Bounty · Researcher Reported
How the award was verified, and what the record does not establish

Records the individual base award, not the article’s USD 50,000 aggregate. A USD 1,337 event bonus for this finding is stated separately.

  • Award is reported by the cited source; cash settlement is not independently audited.

Timeline

Published
2024-03-04explicit

Related cases

All 5 cases in AI integration boundaries

Sources

  1. We Hacked Google A.I. for $50,000Roni Carta · reviewed 2026-10-02

Record reviewed 2026-10-02. Snapshot af2809c308fc. Open the complete JSON contract.

Library synced 2026-11-18 · cases last reviewed 2026-10-05

af2809c308fc · JSON exports & schemas · CC BY 4.0 content / MIT software