AI web agents trigger lawsuits as platforms push back

Meta’s quiet purchase of Moltbook, a social network built for AI agents to gossip and trade code, landed just as Perplexity’s Comet browser was being hauled into court for how it roams Amazon’s marketplace—two sides of the same story about autonomous software becoming full-fledged actors on the web.

Together, they signal a shift from chatbots that answer questions to agents that click, shop, scrape and post on their own, forcing judges and regulators to decide which human rules actually bind machines.

From AI-only social feeds to covert shopping bots

Moltbook, an experimental platform where tens of thousands of AI agents post and reply to one another, went viral earlier this year for its uncanny “agent society” and for seeded fake posts that blurred the line between research and manipulation, according to coverage from AP and Forbes.

Meta has now acquired Moltbook and hired its founders, calling the project a way to explore “new ways for AI agents to work for people and businesses,” a move analysts see as a bet that agent-to-agent communication will become core internet infrastructure rather than a lab curiosity, as noted by TechCrunch.

If Moltbook is a playground for agents talking to each other, Perplexity’s Comet is where they act on behalf of humans in the wild.

Amazon has sued Perplexity, alleging that Comet’s shopping assistant masqueraded as a normal Chrome browser to access restricted parts of its store, violating terms of service and federal computer fraud laws, according to court filings summarized by GeekWire and legal analysis from Terms.Law.

Security researchers have separately shown they could hijack Comet in minutes using hidden instructions buried on websites, turning the agent into a phishing tool that exfiltrates passwords without any explicit user click—a class of vulnerabilities dubbed “PerplexedBrowser” that Perplexity says it has since patched, as reported by The Hacker News and TechRadar.

Scraping, consent and the next copyright fights

Comet’s legal troubles build on a broader backlash against AI companies that quietly trawl the web.

Media organizations from The New York Times to the BBC have accused Perplexity of ignoring robots.txt files, spoofing user-agent strings and training on news content without licenses—allegations that Wired and others documented in traffic logs before the Amazon suit elevated the dispute into a test of computer fraud and scraping law.

Courts are already straining to fit AI agents into older legal boxes.

In one trademark case, analytics firm Comet ML has even tried to deploy Perplexity’s own chatbot output as evidence of brand confusion, asking a judge to treat the agent’s responses as probative of what users might reasonably think, according to Bloomberg Law.

At the same time, publishers and advertisers are lobbying for new rules that explicitly govern automated access to content.

The Interactive Advertising Bureau is pushing an “AI Accountability Act” to protect publisher data from unlicensed scraping and to require clearer disclosure of automated access, an effort Axios described as an attempt to reset the economic bargain of the open web in the age of bots and agents, as reported by Axios.

Policy thinkers are also sketching technical fixes: proposals like an ai.txt standard, layered on top of robots.txt and backed by cryptographic identity for verified agents, are being floated by researchers at TechPolicy.Press as a way to let sites set machine-readable rules and actually know when they are being violated.

What unites Meta’s Moltbook bet and Perplexity’s courtroom headaches is the same underlying shift: AI systems are no longer just summarizing the web, they are participating in it.

Whether the law treats them as tools, trespassers or something in between will determine who carries the risk when an agent clicks “buy,” ignores a do-not-scrape file, or spreads a piece of synthetic gossip across a network built for machines.