Google's Scam Text Lawsuit: The AI Phishing Machine Behind Your Spam

The fake toll notice you got last month. The package delivery text with the suspicious link. The urgent warning that your account was compromised. There is a decent chance all of them came from the same place: a subscription service, sold on Telegram for as little as 88 dollars a week, that uses AI to manufacture scams at industrial scale.

That is the picture painted by a lawsuit Google filed on June 12, 2026 in Manhattan federal court against a China-based cybercrime network it calls the Outsider Enterprise. The complaint, coordinated with the FBI and major US carriers, is the first time Google has sued anyone specifically for misusing its Gemini AI to defraud consumers. The numbers in the filing explain a lot about why your phone has felt under siege.

What the Lawsuit Alleges

According to the complaint and reporting around it, the Outsider Enterprise operated a phishing-as-a-service business. Instead of running scams directly, it sold the tooling: a kit called Outsider, distributed through Telegram, priced at roughly 88 dollars a week or 200 dollars a month, with hundreds of pre-built templates impersonating banks, carriers, toll agencies, and government services.

The scale documented in the filing:

• More than 9,000 fake websites and over 1.5 million fraudulent URLs generated between late 2025 and spring 2026
• Roughly 2.5 million scam texts sent to Android users in a single two-week window in May 2026, drawing about 55,000 spam reports
• Impersonations of Google, YouTube, the US Postal Service, New York's E-ZPass toll system, banks, and phone carriers
• FBI estimates, announced alongside the suit as part of Operation Ghost Hook, of about 3.87 million stolen credit card numbers and roughly 1.9 billion dollars in losses since mid-2023

The mechanism is the important part. Google's complaint describes operators prompting Gemini to write the code for fake reward-redemption and account-verification pages, then loading that code into the phishing platform. Members circulated tutorial videos showing exactly how to do it.

The Skill Barrier Is Gone

For most of the history of phishing, there was a bottleneck: building a convincing fake website and writing believable scam copy required skill. A competent scammer needed a coder, a writer, and time. That kept a ceiling on volume and left fingerprints, like the awkward grammar and broken layouts that trained a generation of users to spot fakes.

AI removed the bottleneck. A low-skill operator with an 88-dollar subscription can now generate thousands of polished phishing pages on demand, each tuned to a different brand, each grammatically clean, each visually convincing. The craft became an assembly line. That, more than any single campaign, is why scam text volume has climbed so sharply, and why the old advice to look for typos stopped working. We covered the broader forces behind the flood in why you get so many spam texts; this lawsuit is the clearest documented example of the machinery.

It is also a preview. The same complaint notes the operation pivoting from package and toll scams to impersonating phone carriers. Templates change in days. Detection rules change in weeks. The economics favor the attacker at every step: sending is nearly free, and one victim in a hundred thousand pays for the month.

The Uncomfortable Loop

There is an irony in the filing that deserves to be named plainly. The scam pages were generated with Google's AI, delivered to phones running Google's operating system, flagged by users of Google's messaging app, and now prosecuted by Google's lawyers.

This is not really a Google failure. Every capable AI model can be misused this way, and Google suing the operators, cooperating with the FBI, and publishing the details is the responsible version of the response. But the loop illustrates something structural: the companies building AI are now in a permanent race against criminal use of their own tools, and the referee, the platform, and the arms dealer are increasingly the same party. Lawsuits punish one network. The kit model survives the lawsuit. Another Telegram channel opens, another subscription starts, and the templates get better.

Why Detection Keeps Losing

The defensive playbook against scam texts is filtering: carriers block known bad numbers, Google and Apple flag suspicious messages, users report and block. All of it helps at the margins, and none of it changes the outcome, for a simple reason: filtering is pattern-matching against an adversary that now regenerates its patterns automatically.

AI-generated scams do not reuse the tells that filters key on. New URLs by the hundreds of thousands. Fresh copy per campaign. Rotating sender numbers. The Verizon 2026 data breach report found mobile channels like text produce substantially higher click rates than email phishing, which means the most effective scam channel is also the one with the weakest structural defenses. SMS was designed in the 1980s with no concept of sender verification at all, a problem we broke down in is SMS end-to-end encrypted.

Detection asks: does this message look like a scam? Against AI-manufactured messages, the answer increasingly looks like everything else.

What Actually Protects You

The practical defenses have not changed, and they still matter. Never tap links in unexpected texts; navigate to the real site yourself. Forward scam texts to 7726 so carrier fraud teams see them. Treat urgency as a red flag. Our complete guide to stopping spam texts walks through every layer, and the field guide to messaging scams catalogs the current patterns, most of which appear in the Outsider templates.

But the honest conclusion from this lawsuit is that individual vigilance is being asked to outpace industrial automation. The structural fix is not better filtering of an open channel. It is a channel that is not open.

A messaging network where every message is automatically verified to come from a real human before it sends, and where you can only be contacted by people you have mutually chosen to connect with, is not a better filter. It removes the input. A phishing kit cannot subscribe its way onto a network that has no path for automated sending and no way to reach strangers. That is the architecture LegitChat is built on: verified-human senders, mutual-consent contact, end-to-end encryption by default. The 88-dollar kit works because SMS and open messengers accept messages from anyone and anything. Change that default, and the assembly line has nowhere to deliver.

The Bottom Line

Google's lawsuit against the Outsider Enterprise is worth reading as a field report on the current state of scam texts: AI-generated, subscription-priced, franchise-operated, and responsible for billions in losses. The suit will hurt one network. It will not change the economics that created it.

Keep the defensive habits. Report what gets through. And recognize what this moment actually is: the point where scam volume decisively outran human-scale detection, and the argument for verified-human messaging stopped being theoretical.

LegitChat launches summer 2026 on iOS and Android. Every message verified human, every contact consented to, encrypted by default. Join the waitlist.