Signal's Whittaker warns AI agents have a 'fatal flaw'

"Did we start trusting agents with our information, or are we just kinda using them because we need to?" That was Meredith Whittaker, President of Signal, opening the "yes" side of an Open to Debate matchup over whether consumer AI agents should be trusted with user data. Opposite her: Neil Chilson, Head of AI Policy at the Abundance Institute, a lawyer and computer scientist. Jeff Berman, CEO of WaitWhat and Co-Host of Masters of Scale, moderated the debate, staged in San Francisco during the Masters of Scale Summit.

Whittaker's case was architectural. Current agents like Meta Muse, Claude Duet and OpenAI Dots, she argued, need "deep permissions" to "act without explicit consent, doing a number of steps," and that same permission surface is what makes them exploitable. "This is prompt injection in LLM-driven agents where they respond to data as if it were an order," she said, calling out "a fatal flaw" in current agentic architectures. Her blunt line on the security tradeoff: "We are eating the seed corn of our cybersecurity when we install these things."

Chilson pushed the other way. "Agents give us agency," he said, "making privacy easier and more convenient for us," and pointed to a concrete consumer win: "Being able to file and register and opt out of data brokers with a six-word prompt is pretty powerful privacy protection." He reframed the question as "safe for what is the real question," and reached for a messaging-era analogy: "We didn't stop messaging because risks existed," he said, arguing instead that the industry "built better platforms."

The transcript published by Open to Debate does not resolve the specific disagreement: whether the fatal flaw Whittaker describes can be engineered out, or whether Chilson's privacy-friendly use cases hold up when the same agent also handles payments and the inbox. Voting on the motion closes November 12, 2026.