Prompt-driven AI extends and transforms surveillance capitalism by increasing the number of ways existing data can be repurposed. This evolution creates greater responsibility for public administrators to use discretion, maintain transparency, and preserve accountability.

Algorithmic and automated systems, such as a formulated Excel sheet or even the cookies collected when visiting a website, have been used for decades to create data points concerning everyday people and their everyday lives. Over time, those data points have increasingly been used to predict behavior in ways that exceed the original purpose of collection. Shoshana Zuboff describes that shift as the idea of behavioral surplus, where data collected beyond what is necessary for the original service can be reused to create prediction products about future behavior (Zuboff, 2019). 

(1) So, what happens when the information an individual willingly relinquishes is interpreted in a manner in which they did not consent? 

Surveillance capitalism, comrade! Zuboff further describes private experience as raw material that can be translated into behavioral data and used to create predictions about futures behavior (2019). Once those datasets exist, however, their usefulness does not remain limited to the organization or purpose that originally created them. For example, in 2026, a regular ol' "Joe" or "Jane" can run a Facebook ad targeting you by using the ticky-tack information that Facebook collected as you created your account (2018). 

But you did not consent to that! Did ya? OHHH yes you did! Worse– you've been doing it for yearsssss.

Internet users accept the terms and conditions of almost every application or website they use because that service provides convenience– whether it is social or material. And, more than likely, that website has informed you that they are collecting your data to do whatever. They are compiling it all into little datasets, on their super secret servers, hidden in their federally protected locations– like Smeagol, they are standing in a corner whispering incoherently about the precious nuggets of information they have gathered from all of the suckers they have tricked into using their services– at least 5 billion, according to the 2026 Mid-Year Data Portal global update report (2026)– so that they may sell it to the highest paying data broker. And, unfortunately, consumers are frequently ignorant of the fact that their personal information may be sold to such brokers for future third-party use.

–––––––––––––––––––––––––––––––––––––––––––––––––

(2.) What Happens When that Existing Data is reused for purposes they were not meant for?

So we, as a society, are becoming more aware that our personal data long outlives its original intent and travels much further than an internet user might expect. Information collected for one purpose does not remain limited to that purpose and once those datasets exist, new data holders may use different systems to ask new questions which is increasingly accomplished by using prompt-driven AI.

–––––––––––––––––––––––––––––––––––––––––––––––––

(3.) What is prompt-driven AI, anyways?

Earlier we mentioned algorithmic and automated systems, or artificial intelligence (AI) and over the last 20 years engineers have trained those systems to interpret and process the human language by feeding them massive amounts of data (Microsoft, 2026). These large language models (LLMs) are primarily prompt-driven, meaning users ideally provide carefully constructed instructions intended to produce informative legible responses for specific tasks or queries (Microsoft, 2026).

Unlike older automated systems that primarily collected, stored, analyzed, and then repurposed behavioral data

(combine SPSS and Excel and you've got yourself a tediously exhilarating new hobby)

prompt-driven AI adds a new interpretative layer to information that already exists. Instead of only organizing the existing data, it can now use that information to generate new characterizations based on the framing of the prompt queried (Microsoft, 2026).

–––––––––––––––––––––––––––––––––––––––––––––––––



(4.) So, How Does prompt-driven ai extend surveillance capitalism?

The mere fact that prompt-driven AI (PDAI) takes datasets meant for one purpose and reinterprets them for another represents an extension of the capitalization of data skimmed from individuals. Zuboff's behavioral surplus theory supports the concern that information can acquire value beyond its original purpose (2019).

Moreover, the original use no longer limits the later questions that can be asked of it, by PDAI. Individuals, often, have limited scope as to what their data has been aggregated for or by whom. Andrew et al., (2023) argue that opaque techniques of data collection make it difficult for individuals to determine whether or not their information is secure and that regulation would enhance accountability surrounding those practices (2023).

Expanded capitalism alone does not explain the full change that PDAI enacts on consumer data, though. PDAI also alters how existing information become interpreted for the user.

(5) Wait... AI can change what data means, too?

YES.

When data is initially collected, collectors and analysts shape the information and categories in which those data points are stored by determining which questions to ask, how to phrase them, and how to code responses. In doing so, societal nuance is already compressed into the data before a prompt is ever written. Once automated PDAI enters the equation, that nuance is altered even further because automated systems manipulate and transpose data without fully understanding the person or situation being represented (Haque, 2015).

But do not forget the prompt writer! The person inputting a query or instructions into a PDAI introduces another layer of framing because the wording of the question influences what the system is asked to notice. In order to fully understand the importance of one's prompt phrasing it would be worth taking a look at how Jurasky et al.(2026) describe the algorithmic processes of LLMs. The thing is that not many LLM users will and because of that the standardization of PDAI prompts may never happen– well... that, free will, poor grammar and laziness.

I digress, the automated nature of LLMs to pull and process tokens ontologically due to the manner in which it was trained, prompts me to take note of the fact that implicit bias can enter not only through the data, but also through the way a question is asked. AI systems are, therefore, capable of producing radical indifference because they reduce lived experience to simple measurable behavior while remaining oblivious to the meaning behind that behavior (Zuboff, 2019).

Once those interpretations begin affecting real people, the question becomes less about technical capability and more about administrative responsibility.

–––––––––––––––––––––––––––––––––––––––––––––––––

(6.)Who is responsible when ai affects someone's life?

AI is slowly infiltrating all avenues of daily life– I mean, I am personally building my own system, built using excel and JSON script, to mitigate the amount of administrative burden I encounter on a daily basis. Such systems can reduce burden and assist with routine decision processes, but it should not replace administrative responsibility, and personnel should exercise discretion whenever an AI-assisted recommendation is capable of affecting another person. Administrative discretion and transparency only become meaningful when everyday citizens can also understand and challenge the process that produced the decision (Andrew et al., 2023).

–––––––––––––––––––––––––––––––––––––––––––––––––

(7.) What should a citizen be allowed to know and who answers for what?

Those affected by decisions made with the assistance of AI systems need enough information to understand how a decision was produced and the appropriate channels through which to challenge those decisions. Opaque administrative practices limit public understanding of how information is used (Andrew et al., 2023), eventually resulting in distrust of the agencies that claim to serve society. At the minimum, effective transparency includes the prompt that was asked, the dataset used to answer it, and contact information for the administrator responsible for reviewing the appeal because if the wording of the prompt can influence the interpretation, then the opportunity to challenge should be made available, as well (Haque, 2015).

PDAI responses can and have led to detrimental consequences– I think of news stories of children committing crimes and self-harm because PDAI affirmed those choices due to a lack in semantic understanding (Hill, 2025). Society has a duty to mitigate such circumstances and each participant may be held accountable at various stages. Especially in cases such as these, the system vendor should answer for the data, ontological categories, and ways in which the system was trained. Pertaining to policy and procedures, administrators should acknowledge responsibility for the prompt used to elicit a certain output. And, the agencies adopting AI systems and employing administrators to use them should be held accountable for their policies and procedures, training and implementation, and oversight. Responsibility should not evaporate simply because several components were enmeshed and contributed to a decision because a lack of discretion, transparency, and accountability can lead to weakened democratic participation.

Individuals affected by policies that have been influenced by AI interpretation ought to have the opportunities to engage administrators in discourse, whether that is agreement, disagreement, or reasoning. Democratic accountability requires public officials and institutions to remain available to the citizens affected by their decisions.