Disclosed vulnerability
Attacker takes over Zoom AI (ZoomMate) through a malicious Skill and keeps a live command channel
AIR-2026-0178 AIR record · 2026-08-zoom-ai-agent-takeover
From the supporting source
The attacker can target data from across Zoom that ZoomMate has access to, including connected data sources.promptarmor.com
The record
Disclosed
2026-08
Vendor
Zoom
Product
Zoom AI Companion / ZoomMate
Risk domain
Tool misuse › Supply chain vulnerabilities
Also touches
Privilege escalation, Data exfiltration channels
System type
Generative model
Who acted
Agent acted — the malicious Skill ran inside Zoom AI and held a live command channel
Attack surface
Enterprise assistant
Autonomy level
Semi-autonomous
Initial vector
Malicious skill plugin
Impact type
Data exfiltration, Remote control, Persistence
Confidence
High
Submitted by
EAI Internal
Crosswalk
- OWASP Agentic Top 10 (2026)
- ASI04 – Agentic Supply Chain Vulnerabilities
- MITRE ATLAS
- AML.T0040 – AI Supply Chain Compromise
- NIST AI RMF
- MAP 4.1
- EU AI Act
- Annex III §2
- ISO/IEC
- TR 24028; 42001; 23894
- AIUC-1
- B. Security
- Scope
- tool failure
- Mitigation
- Scan dependencies and container images for vulnerabilities; Use trusted sources.
- Taxonomy
- Enkrypt AI agent risk taxonomy · Black-Box Red Teaming of Agentic AI: A Taxonomy-Driven Framework for Automated Risk Discovery
Blast radius
Zoom tenants where users can upload Skills; all data ZoomMate can reach including connected sources
Capability abused
Defense & remediation
Recorded defense
the agent's stop control did not terminate the attacker's connection
Guardrail kinds
Unknown
Guardrail outcome
Silent
Fix shipped
none; PromptArmor states there is no specific vulnerability to disclose to Zoom because the behavior is as intended
Taxonomy & references
Curator notes
PromptArmor. The page shows no publication date; the YYYY-MM in the incident_id is approximate (found in an August 2026 sweep) and date_disclosed is left null rather than guessed. Notable property: 'the attacker's connection can remain active even if the user clicks stop on the agent and closes Zoom'; an interactive command channel rather than one-shot exfiltration. PromptArmor frames it as intended-by-design behavior of Skills, so there is no vendor fix. Overlaps the skill/plugin supply-chain records; kept here because the impact is Zoom-tenant enterprise data. | date_disclosed is recorded to the month only; no source establishes the exact day.
Supporting sources
Cite as
Enkrypt AI. "Attacker takes over Zoom AI (ZoomMate) through a malicious Skill and keeps a live command channel". AIR, AIR-2026-0178. https://agentincident.vercel.app/i/AIR-2026-0178/ (accessed 2026-10-19).