Why Judges Are Banning AI in Legal Discovery — and What It Means for Your Privacy

A quiet but meaningful shift is underway in U.S. courtrooms. In the past year, several judges have issued public orders explicitly barring parties from using generative AI tools like ChatGPT or Claude during the discovery phase of litigation. The stated reason: privacy risk. These rulings, recently reported by Bloomberg Law, signal a growing judicial unease with how AI handles sensitive information — and they carry implications far beyond the legal profession.

What happened

Discovery is the pre-trial process where parties exchange relevant documents, emails, and other evidence. It routinely involves highly confidential data — trade secrets, medical records, personal communications, and financial details. To manage the sheer volume, many legal teams turned to AI-assisted document review tools. But a subset of those tools, especially large language models, operate by sending data to third-party servers where it may be stored or used for model training.

Several federal and state judges have now ordered that no generative AI be used in discovery unless the court explicitly permits it. According to Bloomberg Law’s June 5, 2026 report, these orders cite concerns that confidential information could “leak” outside the protective umbrella of the litigation — either through inadvertent disclosure by the AI or because the platform retains input data for its own purposes. The orders typically carve out exceptions for “traditional” AI tools that run entirely on local servers and do not transmit data externally.

Why it matters

For lawyers and self-represented litigants, the immediate consequence is clear: using ChatGPT to summarize a deposition or draft a discovery response could violate a court order and lead to sanctions. But the trend points to a deeper issue with consumer-grade AI.

Many people — including professionals outside the law — now routinely paste sensitive text into publicly available AI chatbots. The assumption is that such interactions are private. They are not. Most consumer AI services log conversations and may use them to improve models. A growing number of companies have updated their privacy policies to note this, but the practice remains poorly understood.

The court orders are effectively judicial recognition of a fact privacy advocates have been pointing out for years: generative AI, as currently designed, is not suitable for handling data covered by attorney-client privilege, HIPAA, or even standard confidentiality agreements. The orders also raise questions about the security of any cloud-based tool used to process personal or proprietary information.

What readers can do

If you are a legal professional, start by reviewing every court order in your case for AI-specific restrictions. Even if none exist now, assume that the default position is moving toward prohibition unless a tool is demonstrably air-gapped or locally hosted. Many firms are now deploying “private” AI instances that run on their own servers and do not connect to the internet. If you are using a cloud-based tool, confirm in writing with the vendor that data will not be used for training or stored beyond the session.

For people outside the legal world, the same caution applies. Do not paste confidential information — medical records, financial account numbers, legal documents — into a free AI chatbot unless you are certain of the platform’s data handling policy. Read the privacy policy, but also consider that policies can change. When in doubt, treat the interaction as public.

More broadly, this trend may accelerate demand for privacy-respecting AI tools. Several vendors now offer “zero-retention” tiers. Some organizations are building their own small models trained on internal data. The court orders could push these alternatives into the mainstream, which would be a net gain for consumer privacy.

Sources

  • Bloomberg Law News, “Judges’ Public AI Bans During Discovery Zero in on Privacy Risk,” June 5, 2026.
  • Individual court orders referenced in the Bloomberg report (specific docket numbers available in the full article).