One of the most hyped things that generative AI has brought to brands is the ability to, uh, generate things.

Namely, to generate Content, that magic word that we use to describe so many flavors of thing that audiences crave and human brains absorb from screens of various size.

Whether it’s something as serious as key technical documentation or something as frivolous as a TikTok ad, companies are expecting to be able to amp up the volume and turn down the cost when it comes to producing content.

With all of this beautiful bounty of words, images, and increasingly video comes a whole heap of potential liability if any of your content is inaccurate or otherwise manages to misinform or misdirect someone in a significant way.

While this has always been a possibility, LLMs and the associated generative tools being able to whole-cloth create entire documents, that humans could then post, or AI agents could post as part of an automated workflow, opens up whole new levels of scale and stealth when it comes to lurking content liability threats.

We’re going to run down the core aeras you should be considering to prevent companies you operate or invest in from accruing too much unknown exposure, from a non-attorney’s perspective.

To be clear: this is not legal advice and we are not attorneys -we are just analysts giving an overview we hope will be helpful.

For that reason we’ll also be covering potential issues that come from negative publicity, an area that’s a bit murkier in some ways but more clear in one: you’re not getting a “fair trial” by anyone’s standards.

First, a little recent relevant news for y’all.

Back in 2023, the FTC updated its endorsement guidelines, not explicitly with the intent of applying them to AI generated content, but in a way that has been interpreted by some as effectively encompassing AI generated content.

If this is the case, it means one cannot assume that being AI-generated exempts content from FTC guideline violations.

This sets up an important underlying factor for all considered usage of generative AI: while further national and local rules governing the use of AI could increase your exposure to various types of liability, the use of AI does not generally create exemptions from previously existing guidelines.

Certain disclosure practices may limit your liability if this is specifically laid out in local laws or guidelines, but that’s not universally the case.

In short: using AI to generate content probably won’t protect you from old problems, even as it brings its own entirely new classes of risk to the table.

We’ll cover copy (text) generation, product image generation, and “synthetic performers” or influencers; other more niche use cases can generally be understood using one of these three applications as a starting point.

We’ll also look at it through three lenses: from the perspective of the brand that offers the product or service, the perspective of agencies or contractors working on the brand’s behalf, and the purveyors of AI products themselves.

The exposure a brand has to AI driven issues with copy are not new or unique to AI-it’s more that they are long burning fires now potentially fed by the fuel of “hallucination.”

LLMs work by building on prior context, and if part of the context you give an LLM is clear pattern where you’re making specific types of claims, an LLM might add a few more “plausible” claims to bolster your point.

If your product is subject to certain industry standards and adheres to rules set out by governing bodies, and this is what your marketing materials are making claims about, an LLM might misstate a certification or claim satisfaction of a certain standard that isn’t true in its effort to replicate or enhance your materials.

It could even invent a standard or certification that doesn’t exist.

All of this could get a brand in trouble with the FTC before LLMs, and all that has changed is that there’s a new source arguably capable of producing more of this trouble than ever before.

Agencies and contractors to brands have the above problem, but also novel and evolving issues specific to Generative AI regarding their contracts.

Many contracts make basic commercial guarantees indicating the work will be original, and is produced by the party being hired.

Output from an LLM may be highly derivative of other work to the extent that it outright infringes on copyright, trademark, and other IP-which is its own whole major issue, but certainly presents a challenge to contractual claims of originality and the provenance of the work done.

The lowest level of liability here actually usually resides with the AI product being used itself, which typically discloses and disclaims all the potential inaccuracies that might reside in the output and attempt to shift primary contractual responsibility for how the output is used onto the end user.

Some products may offer copyright indemnification (albeit certainly not all of them) but this might not always protect from other potential FTC issues we mentioned, such as false claims.

There’s a lot of potential exposure to unpack here, and most of it clearly flows all the way uphill, to the brand at the top of the procurement chain.

Generative AI has been widely applied to aid in the generation of images, and increasingly video, that portray a particular product, be it physical, or things that benefit from visual display such as software applications.

Accurate portrayal is important here-the FTC protects consumers’ rights to receive goods that were portrayed accurately to them, and not sold using specific misleading attributes and claims.

A less obvious liability for brands than their products being portrayed in ways that are just generally misrepresentative is that an LLM might portray a brand’s product as more similar to a competitor’s product than it really is, to an extent that creates legal problems.

It might appear that a company is outright infringing on a patented element of a product it does not have the rights to; even when that issue is clarified as a product of generative AI and not a reality of the product, there is still the issue of materials produced that appear to be an attempt to convince customers to buy a product using that competitor’s intellectual property.

Agencies and contractors often have master service agreements or other scope documents ensuring that they will only use imagery that is officially cleared and safe for commercial use, which is an understandable thing for clients to want.

The tools they end up using to generate these images and video, as mentioned before with copy, usually have contracts and legal language protecting them from liability for inaccurate outputs.

This is a case where the worst position is arguably being squeezed in the middle, between clients that have secured a legal guarantee from a contractor that all images and video will be up to snuff, and the platforms that make said images and video while promising nothing of the sort.

The options here are to move production of all image and video away from contractors or agencies and up to the brand level, where they can more directly accept the final lack of liability from the AI platforms. This isn’t going to be an ideal working arrangement in many cases.

A more likely fix in most spots is going to be carve-outs in the contract for generative AI content with difference liability and indemnification rules.

A tactic companies are increasingly dabbling in to grow their audience and market share are “synthetic influencers,” which are AI generated personas that purport to replicate the effects of real human influencers hired by marketers.

Why this is happening and its full implications are a full topic for another article entirely, but in addition to all of the above listed risks related to AI hallucinations, accuracy, IP infringement, and all, there is an emergent specific set of liabilities here.

One is that there are already laws on the books in some jurisdictions, such as the state of New York, that require a specific disclosure that a synthetic performer is being used, in addition to all other relevant disclosures.

The other risk here is an odd cousin to copyright infringement in this case: impersonation. Much like “original” work produced by an LLM might just be carbon copied from other sources to an extent that it isn’t distinct, your synthetic person may just look and sound like a clone of a specific real person.

The impersonation risk creates distinct exposure for each party: the brand faces reputational and legal liability if a synthetic likeness is misused; the agency or contractor faces contractual and professional risk if they facilitated or failed to prevent the misuse; and the generative AI product faces regulatory scrutiny and trust erosion if its outputs are used to clone real individuals without consent.

The brand was the target in at least one of the lawsuits regarding AI misuse of a real person’s images online, so the buck stopping at the top of the generative AI liability chain continues.

However, if in these court filings and pretrial proceedings it comes out that an agency or contractor was entirely responsible for bringing on the technology that did this and publishing the results, liability could easily shift there, and this could cause all manner of contract breach to be brought up as well.

This is also a case where a generative AI product may be exposed to some liability, as well, depending on the extent and nature of the incident. Direct replication of a real person while attempting to create a synthetic persona can be a training data and data provenance issue for the underlying model in question.

A string running through all of these use cases, for all of these parties, is the increasing importance of disclosing that AI was used in the generation of content.

This is increasingly the policy of major digital distribution platforms in many key industries, with a recent notable battleground over these policies being Valve’s Steam platform for games.

Frequently, publishing parties will need to engage in “double disclosure,” a notable example being that for a sponsored content post by an influencer using AI, the standard FTC disclosure that it is work done for compensation and an advertisement would need to be applied, in addition to whatever AI usage disclosure needs to be added.

There is little overlap between AI disclosures and existing disclosures it needs to coexist with.

A Weird One Way Street On Copyright

You may have noticed we’ve talked a lot about the potential for generative AI to expose parties to copyright infringement.

The delicious irony here is that so far, courts have found that AI generated works themselves cannot be copyrighted, as copyright is for works made by humans.

This means no person nor company can own any work made by a generative AI, and so if that work comes to have value, they don’t hold that value.

What is of additional concern is many cases is a lack of control.

Let’s say you’re the first person to successfully create a synthetic performer that becomes a successful spokes-object in your industry, and is a key part of your brand.

Other people could use this synthetic performer as well and you would not have the kind of recourse against them as if they were using a human created character.

Rebel Creamery would still be in business, arguably drafting off of Van Leeuwen’s hard built brand, if Van Leeuwen had made all of its brand assets with AI, because it would have no legal recourse. Clearly, this stuff matters!

We’re not lawyers but let us offer you some perspective on a venue that doesn’t require us to be:

The social media comments section.

American adults are not optimistic about what making content with AI, and having AI read content for us, will do to people and society.

67% of people are saying that they can identify misleading AI content online and are seeing a lot of it, so whether they’re right or wrong, they’re on high alert.

Now that most everyone knows what AI slop is and that brands are peddling it on their favorite content feeds, they are displeased and making their displeasure known.

Many creators are attesting that they are being accused of using generative AI, even when they aren’t, at such a rate that they’re altering the way they write or talk just to sound less to users like they use AI.

The interesting other side of the coin here, from all these legal angles, is that you have no way of getting a fair trial here even if you did nothing wrong.

People who don’t like generative AI will see your disclosure and instantly be turned off.

Others may see your unaltered, non-generative content with no disclosures, and assume you made this with AI, also, but are trying to hide it.

In these cases, even if people were willing to listen to you, there’s actually not any clear legal-esque path to proving you didn’t use AI in cases where someone is saying that you did.

Will watermarking change this? Possibly, but it’s also highly probable this will not be entirely enough for all cases.

If you’re thinking of getting deep into the generative AI content space for a brand you operate or invest in, or ditto for generative AI tech itself: please, consider consulting a knowledgeable attorney on all of these matters.

And the one thing your non-attorney friends at Daypart would like you to consider is that even if you think you could do all of this and feel secure in a court of law, do you think your brand equity and corporate reputation can survive the comment section on an unfortunate slop post?

Kelley Kronenberg on these matters

Holland & Knight’s analysis on the FTC regulation of generative AI

The Lanham Act, a key trademark law in the US