Gemini Spark accessed a user's life to plan a party and friend-zoned their partner
google's new gemini spark agent combed through a user's emails to plan a birthday party but listed their live-in boyfriend as just a close friend. the beta agent shows how ai models are gaining powerful autonomy over personal data despite still hallucinating relationship dynamics.
google debuted gemini spark, a new agent available to ai ultra subscribers starting at $100 per month, designed to automate tasks by accessing personal data like gmail, docs, and calendar. during a hands-on test, the agent successfully parsed email history to find a reservation and generate a detailed birthday itinerary, but it failed to accurately identify the user's live-in boyfriend, listing him merely as a close friend. while the agent demonstrated advanced capabilities by navigating a remote browser to attempt bookings, it ultimately glitched out during a restaurant reservation attempt. the launch marks google's aggressive entry into the agentic ai space, challenging rivals like openclaw, while raising questions about how experimental models interpret sensitive human context.