The Forum Interstitiale Between Thought and Expression,” by Caroline De Cock

The article is here; the Abstract:

International human rights law protects freedom of thought absolutely and freedom of expression subject to limitations. The relevant established doctrine calls the two sides of that line the forum internum and the forum externum. However, the line was drawn when thinking left no trace, and with current technological developments, it no longer holds.

Within the span of six months in 2026, a United States court ordered that 20 million ChatGPT conversations be provided as evidence under ordinary discovery rules, two federal courts reached opposite conclusions on whether a litigant’s exchanges with a chatbot were protected at all, and the Supreme Court held that location records kept by a third party remain constitutionally protected. Records of where a person’s body had been were protected, while records of how their mind worked were not.

The law currently has no category for the search query, the exploratory prompt, the unfinished draft, or the confession made to a machine. The U.S. answer has been a quarrel about privilege: Courts refuse to grant AI exchanges between users and chatbots protection because the machine is not a professional, like a doctor or a lawyer, and the industry asks for a new privilege because it claims the model behaves like one. Both the courts and the industry assume that protection depends on what the person is using, i.e., what the machine is. This Article argues that it depends on what the person who uses the machine is doing.

It proposes the forum interstitiale, a third zone between the two classical fora, which can protect the thinking process under a standard stronger than ordinary expression law but short of absolute. Every candidate analogy to interactions with AI chatbots in existing law is tested herein, from search queries and confidential dialogue to protected reading, aggregated records, the European law on correspondence and data protection, private diaries and drafts, and neural data. Each fails for the same reason, namely that the protection it offers relates to a relationship, a disclosure, a recording, an addressee, or a data type, and never to the function of the act as thinking.

The interstitial zone this Article proposes only flows upwards. It lifts acts abandoned by the current binary approach into protection and displaces nothing from the absolute core of protected thought. The Article also sets out what acts belong in the liminal zone, five elements of the protection standard, the duties those elements impose on AI providers, and a map of individual AI interactions across the three fora. This classification determines what records of thought can be demanded, kept, filtered, and used against the person who was engaged in the process of thinking.