Signal President warns that agency agency robots against demonstrations threaten the confidentiality of users

• Meredith Whittaker, signal president, said the agent AI has serious security risks to users.
• The agentic AI refers to the bots that can reason and perform tasks for humans without their contribution.
• But having a complete bot of users means giving it access to data trains, said Whittaker.

The president of the Meredith Whittaker signal is skeptical about the AI ​​agent-that is to say AI agents who can perform tasks or make decisions without human contribution.

While some technological titans have presented how useful agents can be useful and launched AI agents so that users can try, Whittaker warned against the risk of confidentiality posed by autonomous agents when speaking during the conference and festivals of SXSW 2025 in Austin on Friday.

“I think there is a real danger we face,” said Whittaker, “in part because what we do is Give these systems so much control that will need to access the data. “”

Whittaker is the president of the Signal Technology non -profit Foundation which manages the application of encrypted signal from end to end known for its digital security.

An AI agent is marketed as a “magic genius bot” that may think of several steps and perform tasks for users so that “your brain can sit in a pot, and you do nothing yourself,” said Whittaker.

For example, she said that the IA agent could perform tasks like finding a concert, reserving tickets and opening an application like Signal To sends friends with the details of concert tickets. But at each stage of this process, the AI ​​agent would access the data that the user may want to keep private, she said.

“He would need access to our browser, an ability to drive this. He would need our credit card information to pay the tickets. He would need access to our calendar, everything we do, everyone we meet. He would need access to open and send this message to our friends,” she said. “It should be able to drive this on the whole of our system with something that looks like a root authorization, accessing each of these databases, probably in clear because there is no model to make it encrypted.”

Whittaker added that an AI agent powerful enough to do so, would “almost” process the data out of the device by sending it to a cloud server and in return.

“So there is a deep problem with security and privacy that haunts this type of threshing Superimpose by joining all of these separate services, crushing their data and doing things like undermining the confidentiality of your signal messages, “she said.

Whittaker is not the only one to worry about the risks posed by the AI ​​agent.

Yoshua Bengio, the Canadian researcher considered one of the AI ​​sponsors, issued a similar warning while speaking to Business Insider at the Davos World Economic Forum in January.

“All catastrophic scenarios with AGA or Superintelligence occur if we have agents,” said Bengio, referring to Artificial general intelligenceThe threshold to which machines can reason as well as humans.

“We could advance our science of safe and capable AI, but we must recognize the risks, to understand scientifically where it comes from, then do technological investment to get there before it is too late, and we build things that can destroy us,” said Bengio.