Key Takeaway
This AI Agents, Autonomous Transactions & Liability Tracker collects AI agent liability cases and earlier automated-transaction cases which may help explain how the law deals with software that takes action outside a conversation. This tracker is a signposting tool. It brings together links and short notes so that readers can find the underlying material more easily. Entries may concern established findings, allegations, suspicions, reported incidents, disputed facts or issues that were never finally decided. Inclusion does not mean that the author accepts any allegation, agrees that AI was used, considers anyone responsible, or suggests that any conduct was unlawful.
The short notes are not a substitute for the full judgment, pleading, report or guidance. Readers should follow the links, check for later developments and make up their own minds. Some entries concern modern agentic AI. Others concern older trading programs, shopping bots, scrapers or automated ordering systems. Those older cases are included as possible legal background, not because they involved today’s generative AI.
Tracker Status: Active/Monitoring
Publication Date: 29 August 2026
Last Verified: 29 August 2026
Latest Case Chronologically:
Latest Legal Article:
Other AI Legal Trackers: AI Law Trackers Hub
Author and Contact: Matthew Lee (Barrister) click here for details.
Ad/Marketing Communication regarding AI Agents, Autonomous Transactions & Liability Tracker
This legal article/report forms part of my ongoing legal commentary on the use of artificial intelligence [TBC] Not legal advice. Not Direct/Public Access. All instructions via clerks at Doughty Street Chambers. This legal article concerns AI agent liability.
The AI Agents, Autonomous Transactions & Liability Tracker
| Date | Case/Matter | Country | Status | Outcome | Summary |
|---|---|---|---|---|---|
| 04/08/2026 | Amazon.com Services, LLC v Perplexity AI, Inc. | United States | Preliminary injunction vacated and matter remanded | Ninth Circuit held Amazon was unlikely to show that Perplexity itself accessed the computers; the user accessed Amazon with the agent as a tool | The Ninth Circuit treated the user, rather than Perplexity, as the actor accessing Amazon on the preliminary record and vacated an injunction against use of the agentic browser tool. |
| 30/03/2026 | The lodge owners v Aston Management International Limited | United Kingdom | Tribunal Hearing | AI not relevant to outcome | In a substantial residential service-charge dispute, the respondent's director had acted as a litigant in person and used an AI agent to prepare his submissions. The Tribunal said that most of the preliminary applications he had sought appeared clearly misconceived on their face. After counsel was instructed, counsel declined to adopt either the submissions or the case citations associated with them. |
| 24/02/2020 | Quoine Pte Ltd v B2C2 Ltd | Singapore | Final appellate judgment | Appeal dismissed on breach of contract and allowed on breach of trust; the automated trading contracts remained valid and enforceable | The court applied ordinary contract principles to automated trades and treated the programmer's knowledge at the time of coding as relevant to the mistake analysis. |
How to Use the Tracker
The main questions are usually what the system did, what the user or organisation authorised it to do, and who the law may treat as responsible for the result.
An entry may concern a clear judicial finding, an allegation, an interim ruling or a reported dispute. It may not yet be known how independent the system really was or whether the disputed action was caused by the software, the user, the provider or several actors together.
What May Be Included
- AI agent liability:
- AI or automated systems placing orders or making purchases;
- software entering websites, accounts or computer systems for a user;
- automated contracts, cancellations, negotiations or messages;
- payments, transfers, refunds or algorithmic trading;
- agents using external tools, databases or APIs;
- questions about authority, attribution, mistake, notice or fraud;
- claims involving access restrictions or website terms; and
- older automated-system cases which may provide useful legal background.
The AI agent liability tracker may include suspected or alleged agent activity where the underlying dispute is public and the uncertainty is clearly stated.
What Is Usually Left Out
Ordinary chatbot answers will not normally be included unless the system also took, or tried to take, an external action with possible legal effect. General copyright and model-training cases are outside the usual scope.
Sources and limitations
Where possible, an entry links to a judgment, order, filed document or official publication. Sometimes the best publicly available source is a report, a party’s account or another secondary source. That difference matters and should be apparent from the entry.
The tracker is not intended to be a complete record of every relevant matter. Links can break, cases develop and descriptions may later need to change. Anyone using an entry for legal, academic or professional work should read the full source, check the current procedural position and look for any appeal, correction, settlement or later decision.
The tracker does not independently determine disputed facts or verify every technical claim made in the linked material.
How to Cite This Tracker
Suggested citation: Matthew Lee, “AI Agents, Autonomous Transactions & Liability Tracker ”, Natural & Artificial Law, first published [insert publication date], last verified [insert date above].
Frequently Asked Questions
What is an AI agent?
Broadly, the term is often used for software which can take steps towards a goal, such as browsing, using tools, accessing an account or placing an order. The amount of human control can vary greatly.
Why does the tracker include older automated cases?
Older cases about trading programs, bots and online ordering may contain useful ideas about contracts, authority, mistake and computer access. They are background authorities and should not be treated as cases about modern generative AI.
Does inclusion mean that the system acted autonomously?
No. A user may have given the original command, approved a step or set detailed limits. The word “agent” can be used loosely, so readers should look at what the system actually did.
Can an AI agent make a contract?
Whether a contract involving an AI agent is legally valid or enforceable will depend on the specific circumstances, including how the AI agent is being used, the parties involved and the applicable law. This is a developing area of law.
Who is liable when an agent causes loss?
Liability will depend on the facts, including how the AI agent was designed, deployed and used, the roles of the parties involved, and the applicable law. Different legal principles may apply in different circumstances.
Does the AI need legal personality?
Whether legal personality is relevant will depend on the legal framework, the role the AI is intended to perform, and the jurisdiction involved.
What if the agent went beyond the user’s instructions?
That may raise questions about authority, mistake, system design and allocation of risk. The answer will depend heavily on the facts, the contract and the relevant jurisdiction.
Can I rely on the tracker to decide who is responsible?
No. It is a set of links and short signposts. Read the full decision or pleading, check the technology described and form your own view.
