Google Sues AI Phishing-as-a-Service Ring: $1.9B Lost [2026] - tech-insider.org

Nadia Dubois July 9, 2026 13 min read Google filed a civil lawsuit on “Outsider Enterprise,” June 12, 2026, using Gemini AI to mass-produce phishing sites. The case, Google v. Does 1-25, landed in the U.S. District Court for the Southern District of New York under docket number 1:26-cv-04982. It marks the first time Google has sued anyone over the abuse of its own Gemini models for fraud. The numbers in the complaint are hard to shrug off. Google says the operation built more than 9,000 fake websites, generated over 1 million fraudulent URLs, and sent 2.5 million scam text messages in a single two-week stretch in May 2026. FBI Cyber Division Assistant Director Brett Leatherman has linked the broader Outsider infrastructure to roughly 3.87 million stolen payment cards and an estimated $1.9 billion in losses dating back to July 2023, according to reporting from eWeek and The Hacker News. This is not just another phishing bust. It is the clearest evidence yet that generative AI has become standard tooling inside cybercrime’s supply chain. The case is now pulling Google, federal law enforcement, three of the biggest US telecom carriers, and Congress in the same direction at the same time, and it raises a question the industry has avoided for years: who is liable when a criminal rents an AI model to build the scam instead of writing the code by hand. · Google · Preferred Sources Don't miss new tech stories on Google Add Tech Insider once in the Google app and our stories appear in your news suggestions. Add Now What Google’s Lawsuit Against Outsider Enterprise Alleges The complaint accuses the Outsider Enterprise of running a subscription phishing-as-a-service business, essentially software-as-a-service repurposed for fraud. Customers paid a recurring fee, logged into a dashboard, and generated fake login pages, fake delivery notices, and fake bank alerts on demand. Google’s filing describes a criminal network that impersonated Google, YouTube, the U.S. Postal Service, toll-payment systems like New York’s E-ZPass, banks, and mobile carriers, then distributed the resulting pages through mass text campaigns aimed mostly at Android users. Google says it worked with outside researchers and law enforcement to trace the operation’s infrastructure, seize phishing domains and admin servers, and pull down a Shopify storefront that the group used to sell stolen data or kit access, according to BankInfoSecurity. The FBI separately seized roughly $100,000 in cryptocurrency from wallets tied to the operation. None of this shuts the group down on its own, which is exactly why Google chose a civil suit: it lets the company pursue injunctions and go after infrastructure that criminal prosecution alone moves too slowly to touch. Inside the Phishing-as-a-Service Kit Sold on Telegram The kit itself, which Google’s complaint refers to as Outsider, was marketed and sold through Telegram channels for as little as $88 a week or $200 a month, per PYMNTS reporting on the unsealed complaint. That price point matters. It is cheaper than most legitimate SaaS marketing tools, and it turns what used to require a skilled web developer into something a low-level scammer can rent for less than a weekly grocery run. Google’s complaint reportedly counts 131 distinct kit variants built off the same core toolset, each tuned to impersonate a different brand or agency. An earlier version of the kit was separately linked to the theft of 36,000 payment cards, PYMNTS reported, suggesting the operation had already been running smaller campaigns well before the scale-up that triggered Google’s lawsuit. That progression, from a modest card-skimming tool to a “Outsider Enterprise,” a phishing-as-a-service kit coached by buyers to generate scam sites using Gemini AI. tooling. What changed this cycle is the tooling itself. Instead of hiring developers to hand-code convincing fake bank portals, the complaint alleges the group turned to Gemini. How Gemini AI Was Allegedly Weaponized for Fraud Google’s filing alleges Outsider Enterprise members used Gemini to generate the code behind phishing landing pages, including customized verification flows and fake “confirm your delivery” screens built to match a target brand’s real site closely enough to pass a glance test. The complaint also alleges Gemini was used to write scam message copy tailored to look like it came from telecom carriers or shipping companies, and that the group produced step-by-step instructions showing less technical criminals how to prompt AI tools to build their own phishing pages, according to the Washington Examiner. Google has called this its first lawsuit over misuse of Gemini for fraud, a distinction that matters because it sets a template. Rather than waiting on criminal prosecutors in a jurisdiction where extradition is unlikely, Google is using civil litigation and infrastructure takedowns to fight the operation directly, and it is doing so while simultaneously trying to show regulators that it polices misuse of its own models. Coding assistants and general-purpose chat models were never designed to generate fraud pages, but the same qualities that make them useful for legitimate web development, fast code generation, natural-language instructions, the ability to imitate a design pattern from a screenshot or URL, apply just as well to a criminal building a fake login page. The Numbers Behind the Case Strip away the legal language and the scale of Outsider Enterprise’s operation becomes the real story. The table below pulls together the core figures from Google’s complaint and the reporting around it. Metric Figure Time Period / Notes Fake websites identified 9,000+ Cumulative, as of the June 2026 filing Fraudulent URLs tracked 1 million+ Nov 2025 – Apr 2026 Scam text messages sent 2.5 million Two-week span, May 2026 Android spam complaints flagged 55,000 Same two-week span, May 2026 Stolen payment cards linked to the network 3.87 million+ Cumulative since July 2023 Estimated financial losses $1.9 billion Cumulative, per FBI Cyber Division Kit subscription price $88/week or $200/month Sold via Telegram Distinct kit variants identified 131 Per Google’s complaint Crypto assets seized by the FBI ~$100,000 June 2026 disruption action Two figures stand out next to each other: a $200-a-month kit subscription on one side, a $1.9 billion cumulative loss estimate on the other. That gap is the entire argument for why phishing-as-a-service, and now AI-assisted phishing-as-a-service, keeps growing. The barrier to entry keeps dropping while the potential payout keeps climbing. FBI, Telecom Carriers and Threat Researchers Join the Response Google did not build this case alone. The FBI’s Cyber Division coordinated on the takedown, seizing phishing domains, admin infrastructure, and the roughly $100,000 in crypto assets mentioned above. Security researchers contributed threat intelligence used to map the network’s infrastructure across multiple hosting providers and domain registrars, a process that typically takes months of correlating registration patterns, hosting overlaps, and payment trails. On the carrier side, AT&T, T-Mobile, and Verizon are now blocking scam traffic tied to the Outsider infrastructure at the network level, filtering messages before they reach a subscriber’s phone rather than relying on the recipient to recognize and report a scam text after the fact. Help Net Security reported that Google’s own messaging protections now intercept more than 10 billion malicious messages a month across its platforms, a figure that gives some sense of how large the background rate of abuse has become industry-wide, independent of any single operation like Outsider. That kind of network-level, multi-party response is relatively new for phishing-as-a-service cases. Earlier takedowns tended to rely on a single law enforcement agency moving against a single kit. This one looks more like a coordinated public-private response, closer to how the security community handles a major botnet disruption than a typical phishing bust. Historical Context: How Outsider Enterprise Compares to Past Takedowns Phishing-as-a-service is not a new business model. What is new is the AI layer. To see how much of a jump Outsider Enterprise represents, it helps to line it up against the phishing kits that came before it. Exact loss figures for older operations vary by source and were often revised months after the fact, so treat the estimates below as the figures reported by researchers at the time each kit was disrupted, not audited totals. Operation Roughly Active / Disrupted Estimated Fake Sites Estimated Losses AI-Generated Content? 16shop ~2018–2021 ~5,000 (est.) ~$500M (est.) No Darcula ~2023–2024 ~3,000+ (est.) ~$300M (est.) No LabHost Disrupted April 2024 ~10,000 (est.) ~$1B (est.) No Outsider Enterprise Disrupted June 2026 9,000+ (reported) $1.9B (reported) Yes, built with Gemini AI LabHost, disrupted in April 2024 in a coordinated international law enforcement operation, was until now the closest comparison: a subscription phishing kit with roughly the same scale of fake infrastructure. Outsider Enterprise’s reported losses run nearly double LabHost’s estimated total, despite covering a similar number of fake sites. The difference is speed and personalization. AI-generated pages can be produced and varied faster than a human template library, which makes each wave of messages harder for spam filters trained on the previous wave’s patterns. Why AI Changes the Economics of Phishing Traditional phishing-as-a-service kits ship with a fixed library of templates: a fake PayPal login, a fake bank portal, a fake shipping notice. Criminals buy the kit, pick a template, and swap in a domain name. Spam filters and brand-protection tools learn to fingerprint those templates over time, which is part of why older kits eventually lose effectiveness and get retired or rebuilt from scratch. Generative AI breaks that fingerprinting advantage. Instead of a fixed template, an operator can prompt a model to generate a fresh variant of a bank’s login page, complete with slightly different HTML structure, different image placement, and different wording, every single time. Google’s complaint alleges exactly this pattern: Gemini used not to design one master template but to keep producing new ones fast enough to outrun detection systems trained on the old versions. That is the real shift in phishing economics. It is not that AI makes any single fake page more convincing than a skilled human developer could make it. It is that AI makes the marginal cost of producing the next 100 unique pages close to zero. Lower marginal cost per page means operators can afford to burn through domains faster, rotate infrastructure more often, and treat detection as a cost of doing business rather than a threat to the whole operation. That is a meaningfully different economic model than the one security teams spent the last decade building defenses against. Market Impact: What This Means for Google and Big AI Platforms There has been no sign of an immediate hit to Google’s stock price tied to the lawsuit, and that is not surprising. A single fraud case, even one this large, rarely moves a company with Google’s scale on its own. The more relevant market impact is regulatory, not financial. Every major AI lab now has to answer a version of the same question from lawmakers, enterprise customers, and insurers: what happens when your model gets used to build a scam, and what did you do to stop it before it made headlines. By filing suit itself, rather than waiting to be named as a defendant in someone else’s case, Google gets to frame the narrative as an AI company actively fighting abuse of its own tools instead of a company that let a “Outsider Enterprise,” the phishing-as-a-service kit itself. That framing carries real weight heading into a period when Congress is actively drafting AI-specific fraud legislation. It also sets a precedent other AI vendors will likely study closely: proactive litigation against downstream abuse, paired with law enforcement partnerships, may become the standard playbook rather than a one-off response. For enterprise buyers evaluating AI vendors, cases like this one are becoming part of the procurement conversation. Security and compliance teams are increasingly asking AI vendors for abuse-detection track records the same way they ask cloud vendors for uptime guarantees. The Congressional Response and Pending AI Fraud Legislation Google is publicly backing several pending bills aimed at scam prevention and cross-border fraud, including the National Strategy for Combating Scams Act, the Strategic Task Force on Scam Prevention Act, and a bill often referred to as the SCAM Act, short for Stopping Cross-border Attacks and Manipulation. PYMNTS also reported that the case has drawn attention from Representatives Fitzpatrick and Harder, whose proposed legislation targets the kind of cross-border scam infrastructure the Outsider case exposed. None of these bills has passed yet, and getting cross-border fraud legislation through Congress has historically been slow, partly because enforcement against operators based in China depends on cooperation that is not guaranteed. Still, the Outsider Enterprise case gives sponsors of these bills a concrete, dollar-figure example to point to in committee, which tends to move legislation further than abstract warnings about future risk. Expect at least one of these bills to get a committee hearing later in 2026, even if full passage takes longer. How Rival AI Platforms Are Handling Misuse Risk Google is not the only AI lab dealing with misuse questions this year. Separately, in June 2026, Anthropic restricted access to two newer models after flagging national-security concerns tied to potential jailbreak methods, a sign that frontier labs across the industry are treating misuse risk as a live operational problem rather than a theoretical one. OpenAI, Microsoft, and Meta all maintain their own abuse-detection and usage-policy enforcement teams, though none has disclosed a case at the scale of Outsider Enterprise. The likely industry response is convergence, not differentiation. AI vendors compete hard on benchmark scores and pricing, but on abuse detection, every major lab has an incentive to match whatever standard the market leader sets, because falling visibly behind on fraud prevention is a reputational risk none of them wants to carry alone. Expect rate-limiting on code-generation requests tied to known phishing patterns, expanded brand-impersonation detection, and more aggressive account suspension for flagged usage patterns across the major platforms over the next several months. What Security Analysts Are Saying About the Case Coverage of the case from outlets including Let’s Data Science, PYMNTS, and BankInfoSecurity consistently frames the Outsider Enterprise takedown as a turning point rather than an isolated incident. The consistent theme across that reporting is that AI-assisted phishing kits remove the last real skill barrier that separated a low-level scammer from a technically capable one, and that detection tools built around static template-matching need to be rebuilt around behavioral and infrastructure signals instead. Analysts covering the fraud-prevention space have also pointed to the criminal network’s business structure as notable in its own right. Outsider Enterprise operated less like a loose hacking crew and more like a software vendor, complete with tiered pricing, product updates in the form of new kit variants, and a sales channel on Telegram. That organizational maturity, paired with AI tooling, is what pushed the case past the scale of prior phishing-as-a-service busts. How Developers and Security Teams Can Respond For engineering and security teams, the practical takeaway from the Outsider case is that URL-based and domain-based detection still matter, even as content generation gets faster and harder to fingerprint. Teams building consumer-facing products can check suspicious links against threat-intelligence feeds such as Google Safe Browsing before rendering them or allowing a redirect. A basic version of that check looks like this: # Check a suspicious URL against Google Safe Browsing (illustrative example) curl -s -X POST "https://safebrowsing.googleapis.com/v4/threatMatches:find?key=YOUR_API_KEY" -H "Content-Type: application/json" -d '{ "client": {"clientId": "your-company", "clientVersion": "1.0"}, "threatInfo": { "threatTypes": ["SOCIAL_ENGINEERING", "MALWARE"], "platformTypes": ["ANY_PLATFORM"], "threatEntryTypes": ["URL"], "threatEntries": [{"url": "https://suspicious-domain-example.com"}] } }' Beyond automated checks, teams that run SMS or push-notification systems should assume brand impersonation attempts will keep varying in wording and structure, since that is precisely the detection gap AI-generated content is built to exploit. Pairing domain reputation checks with rate-limiting on new, low-reputation sending numbers catches more of this traffic than content filtering alone. For readers who simply receive a suspicious text, the FBI recommends forwarding it to 7726 (SPAM) and filing a report with the FBI’s Internet Crime Complaint Center at IC3.gov. Five Predictions for AI-Powered Phishing Through the Rest of 2026 More AI vendors will file their own abuse lawsuits. Google’s decision to sue proactively, rather than wait to be dragged into someone else’s case, gives every other frontier lab a template. Expect at least one more major AI company to file a similar civil suit before the end of 2026. At least one pending scam-prevention bill gets a committee hearing. The dollar figures in the Outsider case, especially the 11 to 18 million accounts (ANTS database breach), give lawmakers a concrete example to cite. That tends to move stalled legislation, even if full passage takes longer. Carrier-level filtering becomes the norm, not the exception. AT&T, T-Mobile, and Verizon blocking Outsider traffic at the network level will likely push other carriers, including regional and prepaid providers, to adopt similar filtering rather than risk becoming the weak link. Phishing-as-a-service pricing splits into tiers. Expect kit operators to start charging a premium for AI-generated, harder-to-fingerprint content, similar to how legitimate SaaS products price basic and premium tiers, while budget kits keep using static templates. Detection vendors pivot toward behavioral signals. With static template fingerprinting losing effectiveness against AI-varied content, expect brand-protection and email-security vendors to invest more heavily in infrastructure and behavioral detection, tracking hosting patterns and sending behavior rather than page content alone. Related Coverage FortiBleed Cracks 86,644 Fortinet Firewalls [2026] New CitrixBleed Flaw: NetScaler Hit in 24 Hours [2026] North Korea Poisons 140 npm AI Packages in 19 Min [2026] SharePoint RCE Exploited: CVSS 8.8, July 4 Deadline [2026] The Gentlemen Ransomware: 483 Victims, 90% Cut [2026] Grok vs ChatGPT vs Gemini: $1.25 vs $5 API [2026] For more coverage of ransomware, breach disclosures, and zero-day exploits, see tech-insider.org’s cybersecurity section. Frequently Asked Questions What is the Outsider Enterprise? Outsider Enterprise is the name Google’s lawsuit uses for a China-based cybercrime network that Google alleges ran a phishing-as-a-service business, selling a kit called Outsider through Telegram for $88 a week or $200 a month. The kit let subscribers generate fake websites impersonating banks, telecoms, and government agencies. How did Outsider Enterprise use Gemini AI? Google’s complaint alleges the group used Gemini to generate the code for phishing landing pages, write scam text message content that mimicked trusted brands, and produce instructions that helped less technical criminals build their own AI-generated phishing pages. How much money did the phishing operation steal? FBI Cyber Division officials have linked the Outsider infrastructure to an estimated 3.87 million stolen credit cards.87 million stolen payment cards, dating back to July 2023. Is Gemini itself unsafe to use because of this lawsuit? No. The lawsuit alleges criminal misuse of Gemini’s code-generation capabilities by a third party, not a flaw that puts ordinary users at risk. Google’s suit is aimed at stopping that misuse, and the company says it is expanding abuse-detection systems in response. What should I do if I get a suspicious text message? The FBI recommends forwarding suspicious texts to 7726 (SPAM) and filing a report at IC3.gov. Avoid clicking links in unexpected delivery, toll, or bank notifications, and verify by going directly to the official app or website instead. How does this compare to previous phishing-as-a-service takedowns? Outsider Enterprise’s reported $1.9 billion loss estimate is nearly double the estimated losses tied to LabHost, the phishing kit disrupted in April 2024, despite a similar number of fake websites. Security researchers attribute the gap to AI-generated content that is faster to produce and harder to fingerprint than static templates. What legal action is Google taking? Google filed a civil lawsuit, Google v. Does 1-25, in the U.S. District Court for the Southern District of New York (case No. 1:26-cv-04982) on June 12, 2026. Civil litigation lets Google pursue injunctions and infrastructure takedowns more quickly than waiting on criminal prosecution alone, particularly against operators believed to be based outside U.S. jurisdiction. Will this lead to new AI fraud regulation? It could accelerate it. Google is publicly backing several pending bills, including the National Strategy for Combating Scams Act and the SCAM Act, and the size of the Outsider case gives lawmakers a concrete example to cite in committee. None of these bills has passed as of this writing. Nadia Dubois AI & INNOVATION EDITOR Nadia Dubois is the AI & Innovation Editor at Tech Insider, where she tracks the rapid evolution of artificial intelligence, from foundation models to real-world enterprise deployment. She previously covered AI and startups for La Tribune and contributed to MIT Technology Review's European coverage. Nadia specializes in generative AI, AI regulation, and the intersection of technology and European industrial policy. She holds a dual degree in Computational Linguistics and Journalism from Sciences Po Paris. View all articles