AI invents software, brands startup a spy: lawsuit hits Palo Alto Networks

What it's about

MeetingTV, a small US startup offering video conferencing and webinars, has sued security firm Koi Security, its four Israeli founders and cybersecurity giant Palo Alto Networks. The lawsuit claims Koi's AI platform "Wings" invented a piece of software that doesn't exist. Based on that, Koi allegedly portrayed MeetingTV as part of a Chinese espionage operation and published it as fact. The Register, Calcalist (CTech), SDxCentral and Startup Fortune, among others, have reported on the case.

The case is being heard in federal court in Southern California. It's the first time an AI hallucination, meaning a fabricated but convincingly worded claim from an AI system, has become the subject of a security-industry lawsuit on this scale.

The "DarkSpectre" report

On December 30, 2025, Koi published a report on its research blog titled "DarkSpectre: Unmasking the Threat Actor Behind 8.8 Million Infected Browsers." In it, the company described an allegedly state-directed Chinese group called DarkSpectre, said to have infected 8.8 million browsers. Part of this was supposedly a campaign called "Zoom Stealer," which harvested data from corporate meetings from 2.2 million users. According to SDxCentral, nearly 30 video conferencing platforms were involved, including Zoom, Google Meet and Microsoft Teams.

In this context, Koi named MeetingTV's browser extension Zoomcorder, a tool for recording meetings. According to The Register, the report called the service the "public-facing front" of the criminal operation, one that "lends legitimacy to the infrastructure while also serving as a revenue stream." Koi also added MeetingTV's domain to its IOC list. IOCs (indicators of compromise) are markers like domains or IP addresses that security systems use to detect an attack. Many firewalls ingest these lists automatically.

The software that doesn't exist

The lawsuit centers on a browser extension called "Twitter X Video Downloader." According to the report, it linked the Zoom Stealer campaign, whose infrastructure the suit says consisted entirely of MeetingTV's servers, to "ShadyPanda," a core piece of the DarkSpectre infrastructure. The complaint states:

"Koi's single-actor theory rested on a fabricated technical 'pivot' – a single piece of software they repeatedly referred to as the 'Twitter X Video Downloader' extension."

According to MeetingTV, this extension doesn't exist. SDxCentral reports it didn't even appear among the extension IDs listed in the report itself. When MeetingTV asked for information about it, Koi allegedly provided none. The plaintiffs suspect Koi's AI hallucinated the software and that nobody checked whether it existed before publication. MeetingTV founder and CEO Michael Robertson told The Register:

"They admit to using AI for their analysis. Maybe a human made all this up? Maybe it was the AI? What's clear is that if the software doesn't exist, even the most basic analysis is impossible – and yet they labeled our URLs, services and software as criminal."

Robertson is a well-known serial entrepreneur. He says Koi never contacted MeetingTV, either before publication or after.

Blocked worldwide

Robertson only learned about the report because security vendors suddenly started blocking his domains. He says he wrote to the companies one by one. Most never replied; one eventually pointed to the Koi report and sent the link. According to SDxCentral, Cloudflare, Fortinet and Cisco Talos, among others, classified MeetingTV domains as malicious or as malware. According to The Register, Verizon and Palo Alto Networks itself are still blocking the startup. The complaint describes a "catastrophic collapse" in online visibility and severe revenue losses.

"When people on the internet can't reach your company, that's a death sentence," Robertson said. "And now all the language models say we work with Chinese cybercriminals. How is that ever going to go away?"

Koi has since removed the references to MeetingTV and Zoomcorder from the report, quietly, according to The Register. According to CTech, an update said further review had found no evidence of a link to Chinese attacks. The original version is still available on the Internet Archive (Wayback Machine). And removing it doesn't lift blocks that other vendors have long since adopted.

The acquisition and a possible motive

Palo Alto Networks announced its acquisition of Koi in February and closed it on April 14, 2026. The company wants the deal to help it grow in "agentic endpoint security," meaning protection for AI agents and automated tools operating inside corporate networks. According to Palo Alto's regulatory filings, the purchase price was 231 million dollars, almost entirely in cash, plus 61 million dollars in stock compensation for the acquired employees. SDxCentral, on the other hand, describes it as a 400-million-dollar deal. That figure can't be directly reconciled with the official filings.

The suit was filed on March 18, before the acquisition closed. On May 13, MeetingTV added Palo Alto as a defendant, arguing that the company took on the liability along with the technology. MeetingTV's lawyers accuse Koi of deliberately using the report as a showcase for Wings during the acquisition talks:

"By fabricating a sensational story about a massive Chinese espionage operation – and presenting Wings as the only solution capable of detecting it – Koi sought to boost its valuation and generate sales leads."

After the deal closed, Robertson wrote directly to Palo Alto CEO Nikesh Arora. He demanded that the report be taken down, that a full correction be published, and that his domains be removed from Palo Alto's blocklist and those of other vendors.

The defense

Palo Alto said it is aware of the lawsuit, that the report was published before the acquisition, that Koi's research reflects a commitment to uncovering threats, and that it expects the matter to be resolved in court. According to CTech, the company has moved to dismiss. Its argument: the report didn't accuse MeetingTV of a crime but analyzed internet infrastructure, and security researchers need protection from defamation suits over mistakes made in good faith and in the public interest:

"The report is security research, not a mudslinging fight between competitors."

Neither Koi nor Palo Alto has publicly explained how Wings produced and verified the disputed findings. On October 2, a federal magistrate judge denied MeetingTV's motion for expedited discovery of Koi documents, but "without prejudice," meaning MeetingTV can file it again later. The case is ongoing.

Our take

Threat intelligence has always had false positives, such as stale indicators or misattributed IP addresses. What's new, according to Startup Fortune, is where the error came from: a language model draws a conclusion, and apparently nobody checks whether the underlying software even exists before the result goes out under the company's name. Because security products ingest IOC lists automatically, an error like this spreads to firewalls worldwide within hours. Undoing it is slow and has to happen vendor by vendor. Robertson's point that chatbots now repeat the false claim highlights another consequence: AI errors can make their way into the training and search data of other AIs.

But keep in mind: so far, this is the plaintiff's side of the story. Whether the link was actually hallucinated or stems from some other error hasn't been proven, and Koi hasn't disclosed its methods. Legally, the question is how far protection for public-interest security research extends when AI output is published without verification. For an industry increasingly relying on AI agents to detect attacks, the outcome of this case will likely be watched closely.