For a long time, the standard toolkit of an antitrust regulator or claimant during a dawn raid or litigation was predictable: Search emails, download server data, and look for smoking-gun text messages or – more recently – WhatsApp chats. But as corporate employees increasingly rely on generative AI tools to draft market strategies, analyze competitors, or optimize commercial operations, an entirely new category of evidentiary data is emerging: AI prompts.
As so often in this space, the US is setting the trend. Recent court cases found AI prompts to be in scope of discovery as part of litigation. This points to a new compliance risk: If an employee interacts with an AI tool, the prompt is not to be confused with an online search or a “private research session”. Rather, it might be evidence that can make or break an entire (antitrust) case.
Here are the top three things companies (and their advisors) should think about:
1. The death of confidentiality: Public AI models and the waiver of privilege
The first critical reality for companies is that typing sensitive queries into a standard, consumer-facing AI model can destroy attorney-client privilege.
In a case in the US (United States v. Heppner), the court ruled that a defendant’s factual and legal queries typed into a commercial AI chatbot were not protected by attorney-client privilege. Because the AI platform’s standard terms of service permitted data logging and model training, the court ruled that the user had no “reasonable expectation of confidentiality.” While the US of course applies a different concept to legal privilege than many European countries, one should not count on different rulings in the EU.
The takeaways:
- Treat public AI as a third party: Feeding sensitive strategic, legal, or commercial information into an unmanaged public AI chatbot can be equivalent to disclosing it to an unprotected third party. In particular when it comes to legal advice, companies should implement a strict policy: Ban the use of public consumer AI tools.
- Mandate closed-loop environments: If employees use AI, they should exclusively use enterprise-grade, closed-loop environments under specific vendor terms that legally prohibit data retention, logging, or model training by the provider.
2. The operational blind spot: Prompts as internal conversations
Even with an enterprise-grade environment, AI prompts created independently by employees seem to generally enjoy no particular legal protection against access by antitrust regulators or claimants. If a pricing, sales, or procurement team inputs competitor data or market-sharing variables into a chatbot, the corresponding logs might be viewed as corporate facts, completely open to regulatory scrutiny during a dawn raid. Unfortunate prompts could serve as a forensic roadmap for enforcers. In a recent commercial dispute in the US (Fortis Advisors LLC v. Krafton, Inc.), AI logs already served as evidence.
The takeaways:
- Adapt internal guidance: Companies should incorporate “prompt-level” guidance into their compliance training and document creation guidelines.
- Educate employees: Employees should be made aware that typing queries into an AI chatbot is structurally identical to writing an internal email or a WhatsApp message.
3. Updating the playbook: Prepare for investigations and litigations
Because AI prompts can be an essential target for regulators and claimants, and because people tend to be very transparent towards the AI of their choice, it is advisable to get ahead, prepare and be ready to mirror how “opposing” forces would approach one’s AI history.
The takeaways:
- Define retention standards: Companies should determine which AI is allowed to be used for what, and define standards for prompt retention, akin to general document retention policies. Where required/advisable, employees should be informed about circumstances in which AI logs might be accessed.
- Establish access frameworks: Together with the internal IT experts, it should be evaluated how AI logs can be accessed, who should “hold the key” and what the access process looks like.
- Expand internal audits: Where companies conduct internal audits including the review of internal documents, they should consider expanding this review to AI prompts.
- Update document holds: Standard document hold instructions to preserve “emails, chats, and files” might no longer be sufficient. Depending on the case, active document holds should explicitly cover internal enterprise AI endpoints, prompt histories, system instructions, and generated outputs. A case in which a company is fined for obstructing an investigation because an employee deleted an AI log is not science fiction anymore.
The bottom line
Antitrust cases will no longer just focus on what employees say to competitors (or the public), they will also focus on what they say to their AI assistants. Companies should prepare their infrastructure and training now, so that AI prompt logs do not become the centerpiece of their next antitrust investigation or litigation.
Picture by Solen Feyissa on Unsplash
Antitrustpolitics.com 