Let's talk tools and terms

Unless you want to pay to be 'patent trolled' you should absolutely read your AI tool's terms of service, and also those of other platforms you use.

I recently received the recommendation for a powerful document editor AI tool. It sounded interesting so I went to have a look at it. The provider’s Terms & Conditions were what caught my attention, which lead me dedicate this week’s post to the issue of T&Cs employed by AI and other platforms.

A. The Tool

The tool that I was initially curious about felt disappointing. It surprised me that - I believe - all it is, is essentially a front-end that serves up a response to an input based on a useful system prompt. Like an elaborate version of ‘create first draft, format it such and such, end on a summary paragraph, make it sound formal’. The tool offers users the choice from multiple AI models, so you are getting access to a few models for one price, but of course you are tied into the corset the platform creates for you.

I find it curious that people would pay a premium for this, in the case at hand over USD300 per year. It seems to me anyone can do exactly the same thing easily and directly in ChatGPT, Claude or Gemini, if not immediately, certainly with a bit of tinkering and reading about prompt engineering. Maybe this is not a fair assessment, because I just haven’t experienced how powerful the platform can be and there is some magic that I don’t see. One thing is certain though: It is unlikely I would use this platform, but please read on. Of course they have built a product and I haven’t, and they are also backed by one of the highest-regard startup incubators, so I guess the joke’s on me.

B. Terms & Conditions

It’s true: old habits die hard. My lawyer mind couldn’t resist so I just had to skim the provider’s Terms & Conditions, and here surprise turned to bewilderment. It is a masterclass on why you absolutely must read the terms of any such AI service you use!

1. Standard stuff

It starts out mostly standard, where you as the user of the platform guarantee to the provider that you have all rights to the information you feed in. You also agree to indemnify the provider if legal problems were to arise, because you weren’t truthful about this guarantee or actually didn’t know. This mechanism is very common and from the provider’s perspective makes sense to protect them: Don’t use our platform and create legal trouble for us. If you do, pay up. No issues up to here.

2. IP rights

The first two paragraphs here are also pretty standard: The platform remains the intellectual property (IP) of the provider, your content you paste into the platform remains yours. So far, so good.

a) Things turn interesting further down though:

You hereby grant to [the provider] an [insert a bunch of legalese to mean: unlimited, worldwide, free of charge, irrevocable, covering every conceivable use-case] right, and license to […] use, copy, reproduce, disclose, sell, resell [and] publish […] your [inputs into the platform]. […]

Ugh - what??? Lets translate this into plain english:

Whatever you are inputting into this platform, you are giving away to the provider for free, forever, worldwide without any chance to change your mind. They can use and exploit your content in any way they like. This applies not only limited to training a model or improving their service offering; no, they can actually commercialise or use your content in any other way!

Lets also translate this into a business rationale:

The platform provider charges you a fee to use their tool. When you use the tool, which you pay roughly USD300 per year for, the platform provider acquires the rights to use and exploit every piece of content you input, in any way they see fit, anywhere in the world. You receive no compensation for this. Essentially you pay for the operating cost of the provider acquiring your content. If this doesn’t qualify as insane I don’t know what will. To me this is the mass market version of patent trolling, except that you also pay for being trolled.

b) Of course the clause doesn’t end here, but continues:

… Additionally, you [guarantee] that you have all rights necessary to grant the rights and licenses [above].

This is an extension of the standard mechanism mentioned above and from the platform provider’s perspective ‘belts and suspenders’. For the user, this guarantee is the source of potential additional trouble via indemnification provision (not duplicated here) that requires the user to hold harmless the platform provider against legal action (as referred to above). The scenario is as follows:

Lets say you input a text where the (IP) is owned by your employer, another business or university, or you aren’t the sole owner of the IP to those inputs (for example you have co-authors). Maybe you wanted to summarise a news-paper article. After your use of the platform, maybe in a few years time the provider then goes to use or monetise this IP in some way (lets recall: anywhere in any way) and gets challenged by the IP’s owner. If this happens the provider can turn to you and force you to pay for the legal fees and damages they are facing, when the actual IP owner enforces their rights against the platform.

To be fair, this ‘guarantee to indemnity’ is a pretty standard mechanism, but it’s worth calling attention to, because of the ludicrously wide set of rights the user grants the platform provider.

c) No silver lining

To be clear, I wonder whether a court would uphold these terms, because of the surprising imbalance they create in favour of the platform provider, but battling this out across state and federal courts in the US and potentially elsewhere will potentially cost millions of dollars, because that battle will be fought over some piece of IP that would be worth it.

From my perspective, this an absolute insane proposition for giving away IP for free and paying for it while running an incalculable legal and financial risk when using an AI service like this.

C. Comparison with OpenAI Terms

For comparison, OpenAI, despite a lot of heat they get on a all kinds of issues, you can’t fault them in this regard: You own all rights to input and output and you allow them to use your content to provide and improve their services, but can opt out of OpenAI using your content to train future models.

This allocation of ownership and rights to content is as you’d expect. In other ways, unrelated to ownership of content, the OpenAI T&Cs also contain some questionable clauses. For example:

Your content. You may provide input to the Services (“Input”), and receive output from the Services based on the Input (“Output”). Input and Output are collectively “Content.” You are responsible for Content, including ensuring that it does not violate any applicable law or these Terms. […]

Should a commercial service distribute the risk of violating laws with its output, possibly because of an infringement of IP rights during the training process, exclusively to the user? This is at the very least surprising, when there is no precise way at for the user to steer or control what the output from the model looks like.

D. Recent LinkedIn changes

If you post on LinkedIn or passively use the platform and haven’t ever bothered to look more closely at their T&Cs now might be a good time: I haven’t seen any official communication in this regard, so I will say: Without informing users LinkedIn changed its T&Cs so that LinkedIn and its affiliates can now use your personal data and content that you create to train their generative AI models, unless you opt out. The setting is toggled on by default (of course).

You can opt out by:

clicking on ‘Me’ in the toolbar at the top;
navigating to Settings;
navigating to Data Privacy; and
at the very end under ‘How LinkedIn uses your data’ toggle the feature to ‘off’.

It should seem strange that this kind of data usage is ‘just slipped in’ without telling anyone. I think the fact that other platforms like Facebook are doing the exact same thing is no defence.

The reality is of course: This is nothing more than AI emboldened iteration of ‘if you don’t pay for the product, you are the product’. I don’t think, however, this should be conciliation to anyone.

E. Conclusion

Please read the T&Cs of the AI and other platforms you use! No-one does and I wonder what wave of law suits we are creating for ourselves without even realising it. Ultimately, the question is of course whether a court would uphold these absurd provisions if a provider like the one at hand were to try and enforce them, so we can’t really say until that day arrives, but do you want to be the one who’s life will turn on finding out?