- The president of the signal, Meredith Whittaker, calls the dangers of AI agents
- Greater use of work tasks means “you are not doing any of that yourself”
- The tools have already raised privacy and security concerns.
The president of the popular Messaging App Signal has warned that AI agents have a significant risk for privacy and security, which is “disturbing” the exaggeration around the agent.
Speaking on SXSW, Meredith Whittaker argued that AI agents are being marketed as a “magical genius” that thinks many steps ahead and complete tasks for users, so “their brain can sit in a bottle and you are not doing any of that yourself.”
But this has a cost, and although the use of AI agents is demonstrating to be popular, Whittaker emphasized that there is a “real danger” with these bots, since they require wide access to user data.
Excessive access
Let’s say he asked an AI agent to reserve a concert for you and your friends: this should be a fairly simple task, but it means that the AI agent would need access to your browser, information about your credit card, your calendar and even your messages to inform friends.
All this means that with only one task, the agent now has access to its financial details, their daily plans and their messages, which could be extremely harmful if the data falls into the wrong hands.
“It would have to be able to drive that through our entire system with something that looks like a root permit, accessing each of those databases, probably clear because there is no model to do that encrypted,” Whittaker explained.
Messaging applications as a signal that have end -to -end encryption (E2EE) would compromise the privacy of users’ messages if they are integrated with AI agents, even if this were just to send text messages to friends or summarize their incoming messages, he said.
“That is almost certain that it is sent to a cloud server where it is being processed and sending back,” added Whittaker. “Therefore, there is a deep problem with the security and privacy that disturb this advertising around the agents, and that finally threatens to break the blood brain barrier between the application layer and the operating system layer by joining all these separate services [and] MUDDING YOUR DATA “
Via Techcrunch
