FPF launches four-part series on agentic AI privacy risks

The Future of Privacy Forum's Center for Artificial Intelligence has launched a four-part virtual roundtable series examining privacy risks linked to agentic AI. Its first session focused on data flows, data minimization, privacy-enhancing technologies and enterprise deployment challenges.
- FPF launched a four-part virtual roundtable series on October 19, 2026.
- The first session examined agentic AI privacy harms, data minimization and PETs.
- A public version 0.1 taxonomy lists 24 agentic privacy risks across five domains.
- Speakers highlighted limits in static permissions, consent and deletion processes.
- The next roundtable is scheduled for December 15.
A framework for agentic AI privacy harms
The first roundtable, titled Agentic Data Flows: Privacy, Data Minimization, and PETs in Agentic Deployment, took place on October 19, 2026. It brought together experts from Working Paper, Google and J.P. Morgan AI Research, with Stacey Gray, FPF's Senior Director for Artificial Intelligence, opening the discussion.
Bennett Hillenbrand, Co-founder and President of Working Paper and an Adjunct Professor at Georgetown University's McCourt School of Public Policy, offered a practical definition of an agent as a large language model that can call tools. Using a multi-agent shopping example, he argued that agents reshape existing privacy risks more than they create entirely new ones. He also said relatively few AI developers were using advanced privacy-enhancing technologies before agents arrived.
Andrew Gruen, Principal Consultant at Working Paper and an FPF Senior Fellow, introduced a taxonomy of 24 agentic privacy risks across five domains. The version 0.1 framework, released for public feedback, covers issues from data collection and aggregation to legacy consent and accountability. FPF says the taxonomy is intended to support benchmarks that test AI systems against these risks before deployment.
Why conventional permissions may fall short
Jaspreet Bhatia, a Staff Privacy Engineer at Google, described the limits of static, yes-or-no permissions for autonomous, probabilistic agents. The challenges discussed included consent fatigue, the lack of standard protocols for agents to negotiate privacy terms with one another, and deletion requests that may not reach agents that have already received a user's data.
Antigoni Polychroniadou, Executive Director at J.P. Morgan AI Research and the AlgoCRYPT Center of Excellence, traced enterprise risks across the application lifecycle. These included agent identity, human approval, execution integrity and liability. She proposed controls such as cryptographically binding agents to the users who authorize them, machine-readable intent, deterministic enforcement and zero-knowledge proofs, with data minimization applied first.
A central theme was that safeguards designed for conventional software, including one-time permissions, up-front consent and free-text instructions, may not map neatly onto agents. Because an agent can share, infer, remember or pass along information step by step, often outside the user's view, controls need to apply to each action. One presenter summarized the concern by saying, “a login is not enough.”
More roundtables will address open questions
The series is part of FPF's Privacy-Enhancing Technologies Research Coordination Network, which is funded by the National Science Foundation and the Department of Energy and presented in collaboration with Working Paper. The network is supported by the U.S. National Science Foundation under Award #2413978 and the Department of Energy under Award #DE-SC0024884.
Future sessions will examine how agents should negotiate privacy terms with one another, how deletion and consent withdrawal can reach downstream agents, how to verify that a person approved an action, who is responsible when an agent makes a mistake, and how privacy-enhancing technologies can be incorporated effectively. FPF says the next roundtable is scheduled for December 15.
For users and organizations, the discussion points to a shift from treating privacy as a one-time permission decision toward controls that follow an agent's actions throughout its operation. The series does not resolve those questions, but it sets out a taxonomy and identifies implementation problems that enterprises may need to address before deploying agentic systems.
Future of Privacy Forum: FPF Launches Roundtable Series on Agentic AI, Privacy, and PETs


