The article is here; the Abstract:
Across international human rights doctrine, freedom of thought is guaranteed in an absolute sense, while freedom of expression remains protected within permissible limits. The traditional framework identifies the two sides of this boundary as the forum internum (the inward domain of thought) and the forum externum (the outward expression). Yet this dividing line was sketched at a time when ideas left no trace, and with today’s advances in technology, it no longer adequately captures reality.
Over a six-month window in 2026, a United States court compelled the production of twenty million ChatGPT conversations as ordinary discovery material; two federal courts issued conflicting rulings on whether a party’s exchanges with a chatbot qualify for protection at all; and the Supreme Court reaffirmed that location data stored by third parties remains constitutionally protected. Physical traces of where a person has been are guarded, while records detailing how their mind operates are not.
Current law lacks a distinct category for a search query, an exploratory prompt, an unfinished draft, or a confession made to a machine. The U.S. debate has centered on privilege: courts deny protection for AI-user exchanges with chatbots on the grounds that the device is not a professional like a physician or attorney, while industry advocates seek a new privilege on the basis that the model behaves in a quasi-professional manner. Both courts and industry alike assume that protection hinges on the tool’s identity—i.e., what the machine is. The present Article contends that protection should hinge on what the human user is doing with the machine.
It introduces the forum interstitiale, a third sphere lying between the two classical fora, designed to safeguard the thinking process under a standard stronger than ordinary expression law but not absolute. The paper tests every conceivable legal analogy for interactions with AI chatbots within existing law—ranging from search queries and confidential dialogues to protected reading, aggregated records, European privacy and correspondence regulations, private diaries and drafts, and neural data. Each contender fails for the same reason: the protection it offers is tied to a relationship, a disclosure, a recording, a recipient, or a data type, and never to the intrinsic function of thinking itself.
The interstitial zone proposed here rises upward. It elevates practices neglected by the current binary approach into protection and leaves untouched the absolute core of protected thought. The article also delineates which acts belong in the liminal zone, identifies five elements of the protection standard, outlines the duties those elements impose on AI providers, and provides a map of individual AI interactions across the three fora. This taxonomy determines which records of thought may be demanded, stored, filtered, and used against the person who engaged in the thinking process.
