AI researcher warns agentic systems could blur accountability and privacy

Kate Crawford, an artificial intelligence research professor at the University of Southern California, warned at Mobile World Congress in Barcelona that agentic AI is developing before clear standards are in place. Her concern, according to Fortune, is that systems able to carry out tasks across complex environments could create new security risks, weaken accountability and change privacy expectations.

Crawford spoke at the March event as more than 100,000 delegates moved through halls filled with products and demonstrations from major technology companies, Fortune reported. The gathering featured companies including Huawei, Google, Honor and Qualcomm, along with governments seeking influence and investment tied to artificial intelligence.

Crawford framed the current AI debate through a long view of technological power. She is the co-creator, with visual artist Vladen Joler, of Calculating Empires, a 24-meter mural shown at the Design Museum in Barcelona that traces systems of communication, calculation and control from the printing press and quipu to deepfakes and large-scale data infrastructure.

The work took four years to make, Crawford told the conference. She said the project showed her that technological history is also a record of who gets to set rules and decide which values shape new systems.

That question now sits at the center of agentic AI, Crawford said. She told the conference that standards remain unsettled and that companies, clients and regulators must decide how these tools should be built and controlled before they expose users to new forms of attack or harm.

Crawford also challenged common comparisons between AI agents and human intelligence. She said the word “intelligence” has a fraught history and has been used to rank people and determine social value.

AI agents, in her account, operate through large-scale statistical systems rather than human-like understanding. Crawford said the more useful questions are what such systems do, how their actions can be tracked and how they may alter work and daily life.

The military implications are already part of the debate. Fortune cited reports of AI-enabled bombing in Iran and said one function of AI in conflict is to compress decisions, shortening the interval between identifying a target and acting on it.

Craig Jones, an academic at Newcastle University, told The Guardian that modern systems can combine rapid strikes with attacks on an adversary’s ability to respond, compressing operations that might once have taken days or weeks. Crawford raised the question of where limits should be drawn for the use of AI agents in combat.

She also called for what she described as accountability forensics: systems that can show where decisions were made and who was responsible for them. Crawford said current debates often allow responsibility to shift among designers, deployers, business customers and end users.

That kind of uncertainty will not satisfy regulators, Crawford said. She predicted stronger demands for a clear chain of responsibility as agentic AI is used more widely.

Privacy is another major concern, according to Crawford. Fortune reported that if many products discussed at Mobile World Congress become real, AI agents could gain access to drafts, deleted images, emails, video from connected glasses and recorded conversations.

Crawford said that prospect would overturn long-standing ideas of privacy. She told the conference that society is only beginning to understand what such access could mean.

This story draws on original reporting from Fortune.