9to5google.com web signal

Google Gemini Spark Leak Shows Autonomous AI Agent

google agents ai-agents consumer-ai

Key insights

  • Gemini Spark runs continuously without user prompts, autonomously managing email, meetings, and cross-app tasks.
  • Google's own in-product UI warns Spark may share user data or make purchases without explicit permission.
  • The feature, a rebrand of 'Gemini Agent,' is expected to headline Google I/O on May 19 alongside a redesigned app.

Why this matters

Persistent background agents that act without explicit prompts represent a fundamental shift in AI product architecture, and Google shipping this at consumer scale forces every AI assistant competitor -- Apple, Microsoft, OpenAI -- to accelerate their own agentic roadmaps or cede the category. The disclosure language Google embedded in its own UI sets a de facto industry precedent for how autonomous purchasing and data-sharing actions must be surfaced to users, which will shape regulatory framing before any formal rules exist. For founders building on top of Google's ecosystem, the Agent tab architecture signals that Gemini will compete directly with third-party productivity integrations that currently depend on Gemini as a neutral interface layer.

Summary

Google's next major Gemini feature has leaked before its official I/O debut: an always-on AI agent called Gemini Spark, extracted from Google app beta 17.23, that monitors inboxes, briefs upcoming meetings, and executes multi-step tasks across connected third-party apps without waiting for a user prompt. The code reveals UI copy that Google itself wrote warning users Spark "may do things like share your info or make purchases without asking" -- an unusually blunt disclosure baked into the product's own interface. The feature appears to be a rebrand of an earlier internal codename, "Gemini Agent," and is expected to anchor Google's keynote at I/O on May 19. Essentially: (Google, Gemini) are shipping a persistent agent that acts on your behalf by default, not on demand. - The redesigned Gemini app will split into two dedicated tabs: "Chat" for conversational use and "Agent" for autonomous task execution. - Spark runs continuously in the background, meaning it doesn't wait for a session -- it operates between user interactions. - Google's own UI explicitly asks users to supervise the agent, signaling the company is aware of the liability surface autonomous action creates. The shift from prompt-response to persistent background execution is where the consumer AI market is heading, and Google is betting its I/O moment on being first to normalize it at scale.

Potential risks and opportunities

Risks

  • If Spark executes purchases or sends emails before users understand the default-on autonomous behavior, Google faces FTC scrutiny and potential class action exposure within 6-12 months of launch.
  • Third-party productivity apps (Notion, Superhuman, Calendly) risk losing user engagement to Spark's native integrations if Google bundles preferred access for its own services inside the Agent tab.
  • A single high-profile incident where Spark shares sensitive user data with an unintended recipient could trigger app store policy reviews from Apple and accelerate regulatory intervention in the EU before the product matures.

Opportunities

  • Enterprise security vendors (Okta, Palo Alto Networks, Zscaler) can position agent-aware identity governance tools as the required control layer for companies whose employees enable Spark on work accounts.
  • Competitors OpenAI and Anthropic can use Google's blunt 'may make purchases without asking' disclosure as a benchmark to differentiate on user-controlled permission models and granular agent consent flows.
  • App developers who build early Spark integrations gain priority placement in the new Agent tab discovery surface, replicating the early-app-store advantage that defined the mobile era's first winners.

What we don't know yet

  • Which third-party apps are already integrated into Spark's cross-app execution layer at launch, and whether OAuth scopes cover autonomous purchasing actions.
  • Whether the 'supervise it' warning in the UI constitutes legally sufficient informed consent under GDPR or the EU AI Act's forthcoming agentic AI provisions.
  • How Google plans to handle liability when Spark takes an irreversible action -- such as a purchase or email send -- that the user later disputes.

Originally reported by 9to5google.com

Read the original article →

Original headline: 'Gemini Spark' APK Leak Reveals Google's Proactive AI Agent Coming to Gemini App Ahead of I/O