Nov 11, 2025

Gil Press writes about technology, entrepreneurs and innovation. Follow Author Share getty The race for AI supremacy is intensifying on multiple fronts, with countries, enterprises, and investors driven by the fear of being left behind. The ultimate expression of data eating the world , AI could end the long suffering of our personal and private data. There is also a very slight chance that AI could resurrect privacy protection and reinvigorate personal data security. The constant collection of data about our lives, both online and offline, prompted me to suggest ten years ago that “the first line of defense against the data deluge drowning our freedoms is to expose, explain and eradicate from public discourse the false tenets of big data religion.” Unfortunately, standing on the shoulders of big data and its “promise,” AI has only increased the motivation to monitor what we do and exploit our data. “If an individual buys a banana at Target, consider how many different entities then get data about that person's purchase of the banana,” Katherine Kirkpatrick Bos, General Counsel at StarkWare, told me recently. “It's disturbing that so much data is being gathered on a day-to-day basis about the activities of an average American.” The AI economy, consisting of ever-growing compute power and language models, AI factories , and sovereign AI , is exacerbating the situation in various ways. “The rules of the games have changed recently with AI because you can fake so many things that you couldn’t before,” says Kirkpatrick Bos. For example, producing a fake passport to open a bank account. ”There’s a huge repository of data everywhere of average people, which is now being fed into machine learning,” says Kirkpatrick Bos. That data is being vastly exposed, given today’s “crisis level of data breaches.” Katherine Kirkpatrick Bos, General Counsel, StarkWare StarkWare MORE FOR YOU There were 2,850 reported data breaches in the U.S. in 2024, up from 1,473 in 2019. AI models not only process vast amounts of data and draw inferences about individuals, but also present a new enterprise cybersecurity challenge. According to the IBM and Ponemon Institute 2025 Cost of a Data Breach Report , 97% of organizations surveyed reported AI-related security incidents and lacked proper AI access controls. 63% of organizations lacked AI governance policies to manage AI or prevent the proliferation of shadow AI. The race for AI supremacy has become personal, with employees competing to get ahead by demonstrating their AI skills and expertise. More often than not, they do it by using AI applications that have not been vetted and approved by the organization (aka “shadow AI”), risking exposing sensitive data. This serious security vulnerability has recently been reinforced by enterprises’ rush to install AI agents. A recent survey from Collibra and The Harris Poll found that 86% of decision makers are confident that AI agents will deliver the expected ROI. But only 48% are establishing formal AI governance policies and frameworks, or providing governance and compliance training to employees (47%). AI agents’ expected benefits are based on their ability to access enterprise data, connect to the external world, and run autonomous actions. These AI agents’ skills also promise to increase “data leakage” and the inadvertent sharing with the world of sensitive data. The availability of personal and private data has always been, at least in part, a consumer choice, preferring convenience over privacy. Most consumers don’t understand what companies are doing with their data and believe they have little to no control over what is being done with it. Still, at least for some consumers, this is starting to change. A recent Deloitte survey found that while 48% of consumers experienced at least one kind of security failure in the past year (up from 34% in 2023), those who trust their technology providers to protect their data spent 50% more on connected devices in the past year than those with low trust in their providers. In the AI economy, changing attitudes about the importance of trust and emerging means for establishing it may become a crucial factor determining the future of data privacy. “Math never lies” says StarkWare and its zero-knowledge proofs enable information verification without revealing the actual data about an individual. Such cryptography-based solutions can be used in various domains to protect private data. For example, Duality ’s homomorphic encryption can be used to combine data sets from different healthcare centers where regulatory and patient privacy concerns prevent data sharing. This allows researchers, providers, and payers to query, analyze, and train AI models on the very best available patient data, without resorting to methods which degrade data quality, such as traditional data de-identification. Recently, Google researchers released a large language model trained with a method known as differential privacy, preventing the model from memorizing examples that occurred only once in its training data, such as a person’s name, address, or phone number. Complex math—and AI—could be used to enhance rather than degrade the state of data privacy. Still, trust-enhancing efforts and the improved security of personal data face an uphill battle in the AI economy. Last week, I received the following message from Facebook: “On December 16, 2025, we’re making changes to our Privacy Policy. Here are some details. We’ll start using your interactions with AIs to personalize your experiences and ads.” “Oftentimes when you talk about privacy, the assumption is that people seeking privacy are wrongdoers or they're seeking privacy to shield themselves or their acts. There's a multitude of legitimate reasons, however, why people want privacy,” says Kirkpatrick Bos. “From the American perspective, I think it's incredibly important to acknowledge that privacy rights are enshrined in the Constitution because the founders wanted individuals to be able to have privacy as a fundamental expression of their morals and values.”