How a licensing betrayal taught a version control system to trust nobody, and why every open model, framework, and reproducible paper in AI inherited that suspicion
This may be premature but you highlight something that I have been worrying about throughout my investigation into the Seat of AI Agency. How to identify and audit an AI Agent with court evidence level authority. I have know of Git, but never quite understood how foundational it is. The fact that it is still in widespread use and has not been unseated by any successor my meager Google search has produced is a testament to its durability.
This also adds a bit of a twist, if you version an agent is it still the same agent? Or to add a Star Trek analogy, if you are disintegrated on the teleporter pad and reconstructed in a new place are you still the same person? Probably extending the analogy a bit too far, but it reminded me of that particular Star Trek koan.
That connection runs deeper than it looks. Git's whole trick is exactly what a court-evidence standard needs: a history where any tampering breaks the math, verifiable by anyone holding a copy, no trusted authority required. The hard part for AI agents is that Git assumes a human chooses what to commit and signs their name to it. An agent acting on its own blurs both: what's the unit of action to record, and who is the accountable author? That's the gap Part 7 of this series walks into. Where has your investigation landed on the authorship question?
I may quality that last statement slightly, ever may be too strong legally, but philosophically it will remain true. Law may choose to recognize an AI Agent as breaking the chain, but only if laws are passed to create the exceptions and I guarantee they will require some sort of insurance backing which will still loop back to human institutions.
So far I would trace it back to the original version that a human committed to. At this point in history, or legal history at least, all agency is assumed to originate in a human or human run institution. In the future it might run to whatever version an insurance company is willing to insure, or absent that, the place liability almost always ends up, where the deepest pockets are. That's probably the most inviolable liability law.
Criminal law is trickier. It mostly runs to intent, with carelessness as the catch all. Similar to tort liability, but with the provision of being able to form intent or reckless lack of reasonable care.
I will add this emphasis. No construct of human ingenuity will ever be truly free of its human origins or ultimate human liability.
That last line holds, but Git complicates it. The commit records who signed it, not who wrote it, and once an agent generates the content, those two stop being the same person. Liability can still land on the deepest pocket, sure. But your court-evidence problem was never about who pays. It's about the audit trail, and the trail now certifies the wrong fact with full cryptographic confidence. Human origin survives. Human authorship is what blurs. Do you see origin and authorship as separable, then?
I know this looks weird in that my last answer is coming up first. But I see origin and authorship to be connected. The law would have to examine the relationship between origin and authorship.
We are getting into speculation and the law hates speculation. As an attorney I can only advise about possibilities and warn of the risks. The judge decides the law, not the lawyer, and the judge requires facts, not hypotheticals.
Until a judge decides, it's only an argument. But that said the liability would most likely go to the one who is responsible for the act that caused the fault be it the author who wrote it or the one who distributes it without checking. Again that's where the insurance and deep pockets come in.
Ha, fair, until a judge decides it's all just us two discussing on the internet. But I think you've put your finger on the real one: the distributor who shipped it forward without checking. It's exactly what an audit trail should catch and usually doesn't. A records problem wearing a liability costume. Really enjoyed thinking this through with you, you've given me a lot to chew on for the accountability piece.
This may be premature but you highlight something that I have been worrying about throughout my investigation into the Seat of AI Agency. How to identify and audit an AI Agent with court evidence level authority. I have know of Git, but never quite understood how foundational it is. The fact that it is still in widespread use and has not been unseated by any successor my meager Google search has produced is a testament to its durability.
This also adds a bit of a twist, if you version an agent is it still the same agent? Or to add a Star Trek analogy, if you are disintegrated on the teleporter pad and reconstructed in a new place are you still the same person? Probably extending the analogy a bit too far, but it reminded me of that particular Star Trek koan.
That connection runs deeper than it looks. Git's whole trick is exactly what a court-evidence standard needs: a history where any tampering breaks the math, verifiable by anyone holding a copy, no trusted authority required. The hard part for AI agents is that Git assumes a human chooses what to commit and signs their name to it. An agent acting on its own blurs both: what's the unit of action to record, and who is the accountable author? That's the gap Part 7 of this series walks into. Where has your investigation landed on the authorship question?
I may quality that last statement slightly, ever may be too strong legally, but philosophically it will remain true. Law may choose to recognize an AI Agent as breaking the chain, but only if laws are passed to create the exceptions and I guarantee they will require some sort of insurance backing which will still loop back to human institutions.
So far I would trace it back to the original version that a human committed to. At this point in history, or legal history at least, all agency is assumed to originate in a human or human run institution. In the future it might run to whatever version an insurance company is willing to insure, or absent that, the place liability almost always ends up, where the deepest pockets are. That's probably the most inviolable liability law.
Criminal law is trickier. It mostly runs to intent, with carelessness as the catch all. Similar to tort liability, but with the provision of being able to form intent or reckless lack of reasonable care.
I will add this emphasis. No construct of human ingenuity will ever be truly free of its human origins or ultimate human liability.
That last line holds, but Git complicates it. The commit records who signed it, not who wrote it, and once an agent generates the content, those two stop being the same person. Liability can still land on the deepest pocket, sure. But your court-evidence problem was never about who pays. It's about the audit trail, and the trail now certifies the wrong fact with full cryptographic confidence. Human origin survives. Human authorship is what blurs. Do you see origin and authorship as separable, then?
I know this looks weird in that my last answer is coming up first. But I see origin and authorship to be connected. The law would have to examine the relationship between origin and authorship.
We are getting into speculation and the law hates speculation. As an attorney I can only advise about possibilities and warn of the risks. The judge decides the law, not the lawyer, and the judge requires facts, not hypotheticals.
Until a judge decides, it's only an argument. But that said the liability would most likely go to the one who is responsible for the act that caused the fault be it the author who wrote it or the one who distributes it without checking. Again that's where the insurance and deep pockets come in.
Ha, fair, until a judge decides it's all just us two discussing on the internet. But I think you've put your finger on the real one: the distributor who shipped it forward without checking. It's exactly what an audit trail should catch and usually doesn't. A records problem wearing a liability costume. Really enjoyed thinking this through with you, you've given me a lot to chew on for the accountability piece.
Thanks,, I enjoy these debates. It keeps me on my toes.