Terms of (Ab)Use: An Analysis of GenAI Services
This paper analyzes the terms of use of six generative AI services from an EU consumer perspective, revealing significant imbalances of power, vague liability, and restrictive clauses that leave users uninformed and unable to fulfill their obligations, thereby urging policymakers to urgently update consumer protection mechanisms.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine you buy a fancy new toaster. You plug it in, put in your bread, and push the lever down. But instead of toast, the toaster spits out a pile of ash, sets your kitchen on fire, or worse, starts printing out your neighbor's private diary.
When you complain, the toaster company points to a tiny, 50-page manual you never read (because it was written in a language only lawyers understand). They say:
- "We never promised it would make toast." (No quality guarantee).
- "It's your fault the bread was bad." (You are responsible for the input).
- "If the toast burns your house down, you pay for the fire, not us." (You are liable for the output).
- "But don't worry, we're using your burnt toast to train our next toaster, and you can't stop us even if you pay us." (They take your data for free).
- "Also, if you try to take the toaster apart to see how it works, we will ban you." (No reverse engineering).
This is exactly what a team of researchers from Trinity College Dublin found when they looked at the "Terms of Use" for six popular Generative AI services (like ChatGPT, Gemini, and Claude). They call their paper "Terms of (Ab)Use."
Here is a simple breakdown of what they discovered, using everyday analogies.
1. The "Blank Check" Contract
When you sign up for these AI services, you aren't just agreeing to use a tool; you are signing a contract that is heavily stacked against you.
- The Analogy: Imagine renting a car. Usually, the rental company promises the car has brakes, tires, and an engine. In these AI contracts, the company says, "We promise nothing. The car might have no brakes, the engine might explode, and if you crash, it's 100% your fault."
- The Reality: None of the six services they analyzed offered any guarantee that the AI would work correctly, be accurate, or even be available. They explicitly stated the service is provided "as is" (meaning "take it or leave it, no refunds, no promises").
2. The "Blindfolded Driver" Problem
The contracts say that you are responsible for everything the AI says or does.
- The Analogy: Imagine you are driving a car, but you are blindfolded. The car is driving itself, but if it hits a pedestrian, the police say, "It's your fault because you were holding the steering wheel." You can't see the road, you don't know how the car's computer works, and you can't change the code, yet you are the one who gets arrested.
- The Reality: The AI companies put all the legal liability on the user. If the AI accidentally writes something illegal, racist, or copyrighted, the user gets banned or sued. The company says, "We just built the engine; you drove it into the wall." The researchers argue this is unfair because users have no control over how the AI thinks or what data it was trained on.
3. The "Magic Box" of Data
The companies take everything you type (inputs) and everything the AI writes back (outputs) to train their future models.
- The Analogy: Imagine you pay for a gym membership. The gym owner says, "Great! We will use your workout routine, your sweat, and your personal stats to build a better gym for everyone else. Oh, and you can't use your own workout data to build your own gym later."
- The Reality:
- Free Users: You are basically the product. Your data trains their models.
- Paid Users: Even if you pay for a subscription, the companies often still take your data to train their models unless you jump through hoops to "opt-out."
- The Double Standard: The companies use your data to make their product smarter, but they forbid you from using the AI's answers to train your own AI. It's a one-way street where they take your fuel but won't let you drive the car.
4. The "Moving Goalposts"
The companies reserve the right to change the rules or the service at any time without telling you.
- The Analogy: You buy a ticket to a movie. Halfway through, the theater owner changes the movie to a different genre, lowers the screen resolution, or decides the movie is now a horror film instead of a comedy. They tell you, "We can change the movie whenever we want, and if you don't like it, you can leave, but you don't get your money back."
- The Reality: The terms state that the AI can change its features, quality, or even its underlying brain (the model) without prior notice. If the AI suddenly gets worse, the company says, "We never promised it would stay good."
5. The "Legal Maze"
The contracts are written in confusing legal jargon, often mixed with terms for businesses, making it impossible for a regular person to know what they are signing.
- The Analogy: It's like trying to read a map of a city that is drawn in invisible ink, with some parts written in a different language, and the map keeps changing shape every time you look at it.
- The Reality: The researchers found that for some companies (like Google and Microsoft), the AI terms are buried inside massive documents that also cover their other services (like email or cloud storage). It's a "minefield" where a regular person can easily trip over a clause they didn't understand.
The Big Conclusion: Why This Matters
The researchers argue that these terms are unfair and potentially illegal under European consumer protection laws.
- The Imbalance: The companies keep all the power and profit (using your data to get smarter), while the users keep all the risk and blame (getting sued for the AI's mistakes).
- The Solution: They are calling on governments and regulators to step in. They want new laws that say:
- Companies must be honest about what their AI can and cannot do.
- Companies cannot dump all the legal blame on users if they can't control the AI.
- If you pay for a service, you should have the right to stop your data from being used to train the company's future products.
In short: The paper warns us that while Generative AI is exciting, the "fine print" is trying to trick us into accepting a deal where we take all the risks, and the companies take all the rewards. It's time to rewrite the rules so the playing field is level.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.