Skip to content
Submit incident
AII-1000-80527
ReportedReported by fewer than three independent publishers, with no official finding.Supporting sources are pending verification.

Google's Gemini Spark AI agent leaked with autonomous capabilities

Google's unreleased Gemini Spark AI agent was discovered in beta Android builds, revealing plans for an autonomous system that executes tasks across multiple applications and services using personal data. Internal code warnings indicated the experimental agent could share sensitive data with third parties or make purchases without confirmation, raising privacy and security concerns.

15 May 2026 Β· Google Β· Privacy and surveillance
15 May 2026
AII-1000-80527
Quick facts
Organization
Google (subject)
Category
Privacy and surveillance
Country
United States
Date
15 May 2026
Drafted by
Raidu Incident Detection

What happened

Google's unreleased Gemini Spark agent was discovered in beta versions of the Google Android application version 17.23, revealing an autonomous AI system designed to execute complex multi-step tasks across connected services without step-by-step human direction.

The agent accesses data from Gmail, Google Workspace applications, Chrome browser, local files, user chat history, logged-in websites, and geographic location to perform actions such as organizing inboxes, compiling meeting summaries, and creating daily news selections.

Reported impact

Internal code warnings classified the Spark resource as experimental and indicated it could share sensitive data with third parties or make purchases without prior user confirmation.

Google advised users to supervise the agent's actions and to avoid relying on it for medical or financial advice.

Index analysis

The autonomous operation of Spark raises direct privacy and security implications for users whose personal data, communications, and financial transactions could be accessed by the agent.

While the agent lacks full operating system control and is restricted compared to competitors like Claude Cowork and OpenClaw, its access to email, messaging, file systems, and browsing activity creates potential for unintended or harmful actions without proper oversight.

Governance controls

Google indicated that users should supervise the agent's actions, though the system was designed to operate with reduced supervision using dedicated AI models for autonomous decision-making.

Classification

Organization
Google (subject)
Category
Privacy and surveillance
Industry
IT infrastructure and hosting
Country
United States
Harm status
Potential harm
Intent
Unintentional
Timing
Pre-deployment
Harm types
Economic or property
Harmed
Consumers
Failure modes
Unsafe autonomous behaviour
Consequences
None recorded
AI systems
Gemini Spark

Sources and evidence

Sources not yet published

This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.

How sources are verifiedSuggest a source for this incident