- ID
- AIA-023
- Category
- AI AI agent incident
- Disclosed
- 2025-08
- Systems
- Gemini for Workspace; Google Home
- Harm type
- Prompt injection / data leak
- Status
- confirmed by vendor
SafeBreach's post reproduces Google's response: Google "acknowledges the research" and says it "deployed multiple layered defenses" and that "These mitigations were validated through extensive internal testing and deployed ahead to all users of the disclosure." That is a vendor acknowledgment of mitigations, not a statement that every issue is fully remediated, so status is not set to fixed.
Summary
SafeBreach reported attacks on Gemini "initiated simply by an attacker sending a Google Calendar invite" containing an indirect prompt injection. It says the demonstrations included controlling home appliances, video-streaming victims and exfiltrating information.
Controls
Controls that could prevent, detect, or support recovery from this kind of failure. Mappings are editorial judgments made by the method, not findings about what any party should have done.
NIST AI RMF
- MEASURE 2.7 Security and resilience are evaluated and documented
- MANAGE 2.4 Mechanisms exist to supersede, disengage, or deactivate the system
NIST SP 800-53
Human-approval (JCR) checkpoint
Sources
- safebreach.com primary link checked 2026-10-31, HTTP 200
Post dated Aug 6, 2025. curl -L returned 403 (Cloudflare challenge) on 2026-10-28; content was read via Firecrawl. Flagged for the weekly link check.
https://www.safebreach.com/blog/invitation-is-all-you-need-hacking-gemini/
Corrections
- Review round 2: the saved SafeBreach copy includes Google's response in its "Vendor Response" section. Status changed from reported to confirmed_by_vendor and status_basis now quotes that response.
To report an error in this entry, see corrections.