Today, Legal Advocates for Safe Science and Technology (LASST) and Encode AI Corporation filed an amicus curiae brief in Amazon.com, Inc. v. Perplexity AI, Inc., partially supporting Amazon’s motion for a preliminary injunction to enjoin Perplexity from allowing Comet, its agentic browser, to access parts of Amazon’s website.
We submitted this brief because the court’s decision may shape how AI agents—autonomous tools that can browse websites, click buttons, fill forms, and complete tasks at superhuman speeds—operate across the internet going forward. As AI agents become more capable, autonomous, and widespread, establishing norms around identification now will help ensure this technology develops in ways that strengthen the internet as a shared, democratic infrastructure.
Why AI Agent Identification Matters
AI agents represent a new category of technology. Unlike simple chatbots, these agents can interpret instructions, navigate the web, and take computer-based actions autonomously. This creates unique risks that we will need to devise technical and policy solutions to address. AI agents can misinterpret user intentions to take harmful actions or be subject to indirect prompt injection attacks, all while operating in ways that make human oversight impossible.
Website operators cannot deploy responsive security measures, like serving modified pages to agents or verifying the AI agent’s authorization to complete a transaction, if they cannot distinguish agent traffic from human visitors. When an AI agent acts in unintended ways, its activity needs to be identifiable and traceable for website operators to conduct a proper forensic analysis, and for the legal system to hold the proper parties responsible.
In addition to explaining these issues, our brief highlights the growing consensus around the importance of making AI activity transparent. NIST’s AI Risk Management Framework identifies transparency as a core characteristic of trustworthy AI and recent state laws require AI systems like chatbots to disclose their nature. And industry leaders are already building solutions: for example, we cite Cloudflare’s Signed Agent program, which verifies web traffic from registered AI agents.
Our Position
LASST and Encode support the aspects of Amazon’s requested relief that would require Perplexity’s AI agents to identify themselves to Amazon’s computer systems. Requiring AI agent identification serves the public interest by enabling website operators to protect themselves and their users, fostering accountability, and encouraging the development of safer, more reliable AI systems.
As AI agents become more capable and widespread, establishing norms around identification now will help ensure this technology develops responsibly.