Wired writer says an AI agent saved money but raised security risks

In short: A WIRED contributing editor says an AI “agent” called Instinct helped with everyday tasks and saved money, but also made costly mistakes and could be risky for privacy and security.

What's going on

AI agents are being sold as helpers that can do things for you, not just chat with you. Think of an agent like a digital assistant that can log into accounts and click buttons on your behalf (like handing someone your phone and asking them to run errands).

In a recent WIRED story, Zoë Schiffer describes testing this kind of tool after trying another agent first. She previously gave Anthropic’s Claude Cowork access to her email and calendar, expecting a very capable assistant. Instead, she says it felt closer to a normal chatbot, just with extra tools.

Schiffer then tried Instinct and reports a mix of real wins and real problems. She says Instinct saved her $550, booked restaurant reservations, and warned her about a phishing scam (a trick message that tries to steal your passwords or money). But she also says it wasted $64. She adds that the tool “might be a security nightmare,” which points to a core issue with agents: they often need deep access to your personal accounts to be useful.

What to watch

More companies are pushing agents that ask for access to email, calendars, and sometimes payment details. The key question is whether these tools can be helpful without needing “the keys to the house,” meaning broad access that could cause damage if something goes wrong.

Source: Wired