At Google's recent I/O developer conference, the company introduced Gemini Spark as an always-on agent that connects to your personal data, completing online tasks and automating daily interactions. My first experience with it was hilarious; the AI not only planned my birthday party in record time but also generated a guest list, which left me speechless—and more than a little embarrassed.

The real kicker? Spark thought my long-term boyfriend was just a close friend and frequent companion, relegating him to a lowly spot on the guest list. I couldn't help feeling like I was living in the '80s again, but not exactly in a good way. The AI even tried booking restaurant reservations for me, only to glitch out and leave me scrambling to call the sushi joint myself.

The post-party suggestions were equally bizarre: Spark exclusively listed gay bars, which made me feel like I was being profiled. However, when I asked for an explanation, it turned out that the AI was simply scanning my Google Workspace history for venues mentioned in emails and travel documents. The system does not make inferences about personal identity, just keywords.

Google is rolling out Gemini Spark as a beta to subscribers of its AI Ultra plan, starting at $100 a month. While it might seem like a dream come true, it's worth considering the potential risks and vulnerabilities associated with such powerful tools. A malicious actor could use these agents to steal private data or manipulate users in ways that could be disastrous.