Amazon and Perplexity have embarked on a battle that could be transcendent in the industry, since involves two cutting-edge technologies in full expansion: AI agents and web browsers that base their operation on artificial intelligence.
The latest chapter is that Amazon.com has sent a harsh cease and desist letter to Perplexity in which it accuses it of violating its terms of service by not identifying itself as an agent and insists that the AI company prevent Comet, its AI browser, from making automated purchases on behalf of users. Comet, like OpenAI’s Atlas and other browsers, includes a broad language model that can automate web browsing and perform actions such as online purchases when users tell it to do so.
Amazon maintains that third-party applications capable of making purchases on behalf of its customers They must request permission from the e-commerce giant before enabling that featureto ensure a positive customer experience: “Third-party apps that act as agents, such as Perplexity’s Comet, have the same obligations, and we have repeatedly requested that Perplexity remove Amazon from the Comet experience, particularly in light of the significantly degraded shopping and customer service experience it provides.”says Bezzos’ signature in a statement.
Another of Amazon’s concerns appears to be that Comet’s AI agent may purchase products other than those suggested by the e-commerce giant’s personalized product recommendations. Something that would go beyond conditions of service and user experience, supporting the thesis of commercial approach to the matter.
Really, a simple solution for these cases, if both companies decided to cooperate (which is not the case), it would mean that Amazon would make its personalization data available to Perplexity, but this would imply an economic compensation for it and for Amazon a rival that would compete with its own native AI agent, Rufus.
Amazon y Perplexity
Comet, Perplexity’s AI browser, currently offers an AI feature that searches and purchases products on various websites, including Amazon, on the user’s behalf. However, Perplexity claims to have received a “aggressive legal threat” from Amazon demanding that it stop allowing its AI assistant to shop for users, something the startup says contradicts Amazon’s values.
The case has its complexity… Should AI agents identify themselves as such or is acting on behalf of users sufficient if they consider it so? And other questions up for debate, doesn’t Amazon want external AI agents on its platform that compete with its own, Rufus? And on the other hand, a few months ago Cloudflare published an investigation that accused Perplexity of extracting data from websites, specifically ignoring requests from sites that blocked AI bots.
Perplexity has responded, also harshly, to the cease and desist that Amazon demands, calling it “harassment”. The AI specialist’s argument is that, since his agent acts on behalf of a human user, he automatically has the same permissions as the latter. This means that you do not need to identify yourself as an agent. For this reason, Perplexity opposes Amazon’s demand and has published an extensive entry on its blog in which it describes the e-commerce giant’s position as “a threat to all Internet users”.
«For the last 50 years, software has been a tool, like a wrench in the hands of the user»they say from Perplexity. “But with the rise of agentive AI, software is also becoming work: an assistant, an employee, an agent”. The startup indicates that large corporations have no right to prevent the user from possessing these tools. «Publishers and corporations have no right to discriminate against users based on the AI they have chosen to represent them. Users must have the right to choose the technologies that represent them..
The debate is complex
Santa Clara University law professor Eric Goldman told The Register that Comet uses Amazon customers’ credentials to make purchases on their behalf and added that “Any service can restrict how a user discloses their credentials. “They could stipulate in their terms of service that credentials cannot be disclosed to third parties.”.
Goldman added that there are reasons why Amazon might not want to do that, as users may find it useful to have services act as intermediaries on their behalf. “For example, many financial applications request banking credentials to collect user data and potentially conduct transactions on their behalf.”he explains. Banks could block this access. But if they do, they would be excluding a segment of users who value this type of access above any other.
If Amazon prohibited the disclosure of credentials under its terms of service, the user would be in breach of that agreement by sharing them with Comet, and Perplexity could face additional liability for acting on credentials it should not have. The professor added that Amazon could choose to block Comet specifically, if Amazon could reliably identify the browser; a challenge that has baffled companies that have tried to block AI companies’ trackers.
Goldman said it’s unclear whether Amazon could win a computer fraud case if Perplexity refuses to negotiate or change its behavior. “The courts are in chaos on this issue, especially after the US Supreme Court’s decision. in the Van Buren case. Today, no one knows for sure whether scraping (the extraction of information from websites) “it’s legal”.
Doubts for the future, like everything in AI
The case of Amazon and Perplexity is an example of another of the great problems of artificial intelligence technologies and directly involves one of its pillars, AI agents in their relationship and ability to interact with web pages, search engines, and browsers and with all the services that the Internet offers, including electronic commerce.
Specialists agree that, for the legal system, Distinguishing between web browsing, data mining and AI access is going to be extremely difficult. The Californian professor explained that the dispute reminds him of contractual battles between cable television companies and broadcasters. “It’s like those things where both parties have to suffer before realizing that it is better for them to work together”.
And beyond other questions. How much responsibility do AI agents have for everything they are capable of doing? Are the virtual users of the physical people who approve them or something else? The debate will continue.
