Report from Courthouse News
In Brief – US District Judge Maxine Chesney has granted Amazon a preliminary injunction blocking the use of an AI-powered shopping assistant created by Perplexity AI on Amazon’s website. Chesney determined that Amazon is likely to succeed in its claim that Perplexity’s Comet tool violates both the US Computer Fraud and Abuse Act and California’s Comprehensive Computer Data Access and Fraud Act. According to her ruling, Amazon presented strong evidence that Comet accesses users’ password-protected Amazon accounts with the user’s permission but without Amazon’s authorization. The tool then transmits private account information to Perplexity’s servers to complete shopping-related tasks. The judge ruled that the balance of hardships favored Amazon, noting that while the injunction may harm Perplexity’s effort to gain a first-mover advantage in AI-powered shopping tools, Comet can still operate on other websites. Amazon’s lawsuit alleges that Comet disguises its activity by identifying itself as a Google Chrome browser instead of an automated AI agent and argues this lack of transparency violates its policies and poses cybersecurity risks. Perplexity, which is appealing the order, counters that the lawsuit is an attempt by Amazon to block a competing AI shopping assistant that could undermine its advertising-driven marketplace.
Context – Amazon is the second big platform to sue Perplexity, the self-described Generative AI (GAI) “answer engine”, for violating federal law by circumventing access restrictions. Reddit’s lawsuit alleges that the company engages in “industrial scale” evasion of access controls on Google to scrape Reddit content off that platform and display it on Perplexity. Google has not sued Perplexity but has sued its data-scraping partners. The Amazon suit, and Judge Chesney’s initial ruling, is one of the first that may set parameters around AI assistants and AI agents. In short, if assistants and agents require authorization from both the users and platforms they visit, it may put a break on their now-rapid evolution. Security experts are very concerned about whether current cybersecurity protocols and methods are fit for AI agent operations.