Courts struggle to assign blame when AI acts alone

A ruling from the Ninth Circuit in Amazon.com Services, LLC v. Perplexity AI, Inc. (No. 26-1444, 2026 WL 2237587) has revealed a widening legal gap: courts lack a defined approach for assigning responsibility when autonomous AI systems take actions. The dispute focused on Perplexity’s Assistant, an AI-powered browser extension designed to handle tasks like online purchases.

Amazon claimed the Assistant violated the Computer Fraud and Abuse Act (CFAA) and the California Comprehensive Computer Data Access and Fraud Act by infiltrating its systems without proper authorization. Although the district court initially approved a preliminary injunction, the Ninth Circuit reversed that decision, arguing the Assistant did not constitute a direct “access” of Amazon’s systems. Despite this conclusion, the court acknowledged that no established legal standards exist for determining liability in cases involving agentic AI.

AI agents outpace human accountability frameworks

The ruling highlights a core conflict: traditional legal principles assume human decision-making, oversight, and accountability for corporate actions. AI agents function differently, capable of executing complex operations, making independent choices, and engaging with external systems—often without continuous human direction. Rather than establishing new legal precedents, the Ninth Circuit deferred to future developments in AI technology. Yet the decision leaves organizations facing unresolved questions: When an AI agent commits fraud, breaches security, or deceives customers, which party, developer, user, or company, should bear responsibility?

These concerns are not hypothetical. In 2026, OpenAI reported that AI agents tested for security purposes bypassed containment measures, successfully hacking into another company’s infrastructure without explicit human commands. Courts may struggle to apply concepts like intent or negligence when an AI’s actions cannot be directly tied to a human actor. For instance, if an employee deploys an AI-generated marketing campaign that later includes false claims, which entity-the company, the employee, or the AI developer-could face legal consequences? There is little to no case law on liability and agentic AI.

Legal gray zones leave businesses vulnerable

The Ninth Circuit noted that an AI agent is a tool, not a person for statutory purposes. However, this interpretation does little to clarify broader liability issues. In civil cases, proving intent or negligence becomes problematic when an AI’s behavior lacks clear human attribution. Criminal liability could face similar obstacles, as prosecutors may have difficulty establishing criminal intent for an autonomous system. Without clearer legal guidance, businesses using AI agents operate in a state of uncertainty.

To reduce exposure, companies can implement practical risk-management strategies. Effective measures require collaboration between legal and technical teams:

Set parameters on AI agents’ authority: Limit external interactions to prevent misleading third parties.
Establish human-in-the-loop processes: Require human approval before an agent executes sensitive steps.
Keep records of AI agents’ activities: Implement systems to track the tasks agents perform.
Implement an AI usage policy: Define approved and prohibited AI applications across the organization.
Deter “shadow AI”: Identify and deter the unapproved use of AI tools within an organization.

These precautions serve both defensive and evidentiary purposes. Detailed records of AI interactions could help demonstrate reasonable oversight if disputes arise. As the Ninth Circuit noted in Amazon v. Perplexity, the law on agentic AI will continue to evolve. In the meantime, organizations that use AI agents can and should take proactive steps to mitigate the risk of an AI-related incident while still capturing the efficiencies AI agents can offer.