Lorem

Probably even the chatbot ChatGPT[1] would be “surprised” to see how many new articles, essays, blogposts, personal posts and videos discuss issues around AI, and specifically, generative AI.

In this blogpost we’re not going to try to fully explore the various legal aspects of AI. We want to focus on one narrow intersection: between generative AI tools and competition law. We’ll look at – or rather ask some initial questions – how each main area of competition law could be relevant to the business conduct of ChatGPT (or rather OpenAI Inc., the company behind the chatbot) and similar generative AI tools.[2]

Anti-competitive agreements

The prohibition of anti-competitive agreements (in other words, the cartel prohibition) applies to all undertakings. This includes companies that provide generative AI services, like ChatGPT. They’re innovative services, but the companies involved (including managers and employees) are prohibited from engaging in any collusive conduct (eg concluding agreements on the fixing of prices, sharing markets or exchanging sensitive information).

Are there any specialities in the generative AI business in this respect? The following points merit special attention:

  • Generative collusion: How and to what extent can competing generative AI tools “use” each other, learn from each other and engage in some kind of “interaction”? If one chatbot can access the public business conditions (including prices and terms) applicable for a competing chatbot, it can then suggest (or make) changes to its own business conditions. This can be seen as autonomous adaptation to market conditions, which should not raise concerns. But if this is done by most or all chatbots on the market, could the chatbots realise the advantages of cooperation as opposed to competition and come up with the same general terms and conditions or prices?
  • Platform collusion: A chatbot could provide a platform for collusion for players in an entirely different market, especially if it can collect and disseminate data.[3] For example, competitors in an industry could provide the chatbot with their current pricing, output or other sensitive data. Then it could get data about others by putting relevant questions to the chatbot. Is the chatbot liable – just like a well-known consultancy firm once was for organising secret physical meetings for the worldwide hydrogen peroxide cartel – as a contributor under EU competition law?[4]
  • Cross-shareholdings / cross-directorships: Many Big Tech companies and other investors have shares in various generative AI companies. These companies – which may be direct competitors – need to pay particular attention to erecting internal and external barriers to prevent the exchange of sensitive information (eg business plans, current / future pricing data). These compliance measures are essential to enable fully autonomous decision-making at each generative AI company and to secure a fully competitive landscape.
  • Compliance: ChatGPT could also be an incredibly useful tool to foster competition law compliance. For example, by generating or reviewing simpler compliance documents for companies (eg drafting a new manual to handle dawn raids by competition authorities). A competition law expert should closely supervise this to ensure accuracy and completeness. Using AI for compliance is not a new phenomenon: DLA Piper has already been using its own AI tool, Aiscension, to quickly spot evidence of anti-competitive behaviour in large batches of documents (such as entire email boxes).

Abuse of dominance

Competition law also prohibits abuse of a dominant position. Could ChatGPT or other similar service providers be regarded as dominant, either individually or collectively?

ChatGPT is an extremely popular tool. It had more than 100 million users in February 2023.[5] There are also major competitors to OpenAI’s new system – Google Bard, Facebook’s RoBERTa or and also innovative startups like Caldera Labs Inc.’s Cohere or AI21Labs’ WordTune.

  • To provide a more nuanced view on possible abuse issues, we’d have to review the precise arena of competition for generative AI services, ie the relevant product and geographic market. ChatGPT itself answered this question:

ZO: What is the relevant market for ChatGPT under European competition law?

ChatGPT: As an artificial language model, ChatGPT could potentially be considered a provider of digital services, specifically in the field of natural language processing and generation. Therefore, the relevant market for ChatGPT under European competition law could be the market for digital services related to natural language processing and generation.

However, it is important to note that the relevant market for ChatGPT would ultimately depend on the specific products or services it offers, as well as the relevant geographic market in which it operates. Additionally, market definition is a fact-specific inquiry that requires a detailed analysis of the relevant product and geographic markets, as well as competition dynamics, which would need to be carried out by a competition law expert.

ChatGPT – although with clever disclaimers[6] – appears to position itself as a provider of digital services, putting itself on the narrower market for natural language processing and generation. It’s been argued that ChatGPT could cause disruptive competition on a wider scale. It may put competitive pressure on the market of general internet search engines (such as Google Search).

For consumers it could be a substitute for – and may be even superior to – general search services.[7]

  • Although the market for natural language processing and generation is innovative, there are significant barriers to entry. For example, the exponentially increasing costs of training the large language models (LLMs) used by generative AI companies.[8]  
  • A possible indication of dominance could be that generative AI services may also fall under a special regulation ­– Regulation 2022/1925/EU, known as the Digital Markets Act. It’s even possible the European Commission will classify companies providing generative AI services as “gatekeepers.” It could be argued they provide “core platform services” (like virtual assistants and search engines).

This shows the issue of possible dominance is certainly too early to call. But the relevant market players will have to be watched closely to ensure they remain open, innovative and competitive.[9]

Merger control

Major tech players are already investing in various generative AI products, including ChatGPT. If these investments reach the level of control defined in most merger control regimes (eg majority of voting rights or a veto right on important business decisions), these transactions (as concentrations) could be scrutinised by antitrust authorities.

We’re not aware of any transactions notified to a major competition authority where generative AI was at the forefront of the competition assessment. But sooner or later merger control will also play a role in this field of the economy. Questions about the relevant market, and the possible concentration of the generative AI sector, will likely be highly important.

Unfair competition

A number of domestic laws (eg in Germany, Hungary) include overall prohibitions on unfair competition, outside the scope of the classic antitrust world. For example, the Hungarian Competition Act prohibits unfair competition in general. It also includes specific prohibitions such as the one relating to passing off (preventing one trader from misrepresenting goods or services as being the goods and services of another (competing) trader). And one on breaching business secrets.

  • One issue with generative AI is that it’s extremely easy to produce text – intentionally[10] or unintentionally – that resembles that of other companies or brands. ChatGPT (or, more plausibly, ChatGPT’s sister product DALL-E, the image generator) could be seen as a direct competitor to the artists or designers whose style it can mimic.

When ChatGPT produces a new text in the distinctive style of an artist, and if the text is then used by a company (eg as a slogan for a new commercial), it could harm the artist and cause a significant loss of business. This could be described as unfair competition in the form of passing off. Importantly, this prohibition could also apply even if, from an IP perspective, there’s no copyright infringement (eg because under copyright law, as opposed to specific works of art, style and stylistic elements are much more difficult to protect).

  • Breach of business secrets is also an issue. Under EU law, trade secrets are defined by Directive 2016/943/EU as information that’s:
  • secret;
  • has commercial value because it is secret; and
  • has been subject to reasonable steps under the circumstances, by the person lawfully in control of the information, to keep it secret.

EU law also prohibits the unlawful acquisition, use and disclosure of trade secrets. What if ChatGPT learned of a trade secret during its training (eg information made inadvertently available on the internet)? What if ChatGPT learned of a trade secret from new data being inputted into it (eg if a lawyer copy-pastes a client’s contract to have it reviewed for specific clauses)?[11] What happens if ChatGPT then reveals the newly learned information or secrets to an interested third party? Arguably, such an “action” by ChatGPT could be a breach of the prohibitions and could constitute unfair competition.


[1] The term is an abbreviation of Chat Generative Pre-trained Transformer. According to OpenAI, ChatGPT is a “model […] which interacts in a conversational way. The dialogue format makes it possible for ChatGPT to answer follow-up questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests.” (https://openai.com/blog/chatgpt)

[2] According to public sources, the OpenAI group consists of OpenAI LP and its subsidiary, OpenAI Inc. For further details, see https://fourweekmba.com/who-owns-openai/

[3] This, however, probably cannot be achieved with the current version of ChatGPT, which has a clear limitation in this respect as it only has “limited knowledge of world and events after 2021.”

[4] See the European Court’s C-194/14 P AC Treuhand II judgment.

[5] https://www.reuters.com/technology/chatgpt-sets-record-fastest-growing-user-base-analyst-note-2023-02-01/

[6] Possibly, admitting complexity of a question is a human trait in itself – see the King’s Counsel column in the Sunday Times: https://www.thetimes.co.uk/article/kings-counsel-march-2-2023-nh25q6d8p

[7] On the competition between Google and Microsoft in this respect and whether such competition is indeed only “one prompt away,” see https://www.networklawreview.org/bing-chatgpt/

[8] https://twitter.com/tomgoldsteincs/status/1544370885213175808?s=20&t=JiautEdOA_WyP3UZ4G25bg

[9] Interoperability / third party access is already an issue that merited discussion. ChatGPT has very recently allowed third party developers to integrate ChatGPT into their own apps and products through ChatGPT’s application programming interface (API). This step is already hailed as the possibility to let the “gold rush” to begin in this field (https://www.wired.co.uk/article/chatgpt-api-ai-gold-rush). 

[10] Eg as a result of a specific instruction to use the style of a certain author or brand.

[11] Some companies show an automatic warning page to employees when they click on the website of generative AI products on company computers stating that they must not enter any confidential information relating to the company or its clients.