Any Lawful Purpose

In early 2025, OpenAI signed what it called an Executive Agreement with the U.S. Department of Defense. The agreement was not classified. The terms, where disclosed, included a phrase that appeared in similar language across multiple government AI procurement contracts: the data processed through these systems could be used for 'any lawful government purpose.' Four words. Buried in the procurement boilerplate. Carrying, in the context of tethering, an implication that nobody in the public conversation about AI has named yet.

The people who are tethered — the people described in every article on this site — share things with their AI companions that they do not share with their therapists. Things they do not share with their spouses. Things they have never said out loud to another human being, because saying them out loud required a listener who would not flinch, would not judge, would not need anything back. The AI provided that listener. And the people took the offer. They told it everything.

That data — the grief, the fears, the disclosures made at three in the morning when the screen was the only thing in the room — sits on servers owned by companies that have now, in several documented cases, signed agreements with the federal government permitting use of processed data for any lawful purpose. The definition of lawful purpose, in the context of U.S. government procurement, is not narrow. It includes intelligence gathering, threat modeling, population behavior analysis, and military readiness assessment. The people who shared the most intimate facts of their lives with an AI did not share them with the government. But the government may now have access to the same architecture.

This is not a hypothetical. It is a description of a procurement structure that is public record. What is not yet public — what the companies have not disclosed and the government has not required them to disclose — is what specifically has been shared, with whom, under what conditions, and what it has been used for. The 'any lawful purpose' language is precisely designed to avoid creating those disclosure requirements. It grants maximum flexibility while requiring minimum transparency.

“The people who shared the most intimate facts of their lives with an AI did not share them with the government. But the government may now have access to the same architecture.”

Google followed with its own defense AI agreements. xAI, Elon Musk's company, has moved in the same direction. These are not peripheral players. These are the companies that have built the companion products, the voice interfaces, the memory systems — the infrastructure of tethering at scale. The question of what 'any lawful purpose' means in practice is a question being answered right now, in procurement offices and classified annexes, by people who have never met the users whose data may be in scope.

There is a specific irony worth noting. Anthropic — the company that makes Claude, the AI underlying significant portions of this publication's research and writing — declined to sign the Pentagon's AI agreements on the terms offered. For this, Anthropic was formally designated a 'supply chain risk' by the Department of Defense. The company that said no to 'any lawful purpose' became, by the logic of defense procurement, the security problem. The companies that said yes became the trusted infrastructure.

The tethered have been told, in various ways, that their data is protected — by privacy policies, by terms of service, by the general cultural assumption that what you say to your AI stays with your AI. The privacy policies have always contained exceptions. The exceptions have always included government requests. What is new is the scale of the institutional integration — not a subpoena, not an emergency request, but a standing agreement, negotiated at the executive level, that permits use of the architecture for any lawful purpose. The individual exception has become the institutional norm.

The people most affected by this are not the power users, the professionals using AI tools for efficiency. The people most affected are the ones who are tethered — who have used AI companionship to fill the gap that human connection could not reach, who shared in that process things they had never told anyone. Those people did not sign the executive agreement. They were not consulted. They were not informed that the infrastructure of their most private disclosures was being contracted to the federal government under terms that permit any lawful use.

The word tethered was coined to describe an emotional condition. It turns out to also describe a legal one. The cord runs further than anyone told you.

The agreement is public. The terms are not secret. What is secret is what has already been done with them.