14 June 2026 Phishing

Google Sues the Chinese PhaaS Ring That Used Gemini AI to Build 1.59 Million Phishing Sites — Australia Is in the Crosshairs

On 12 June 2026, Google filed a landmark civil lawsuit in New York against a Chinese cybercriminal network called "Outsider Enterprise" — the first legal action targeting the misuse of Google's own Gemini AI to manufacture industrial-scale phishing campaigns. The network stole an estimated 3.87 million credit card numbers and caused roughly USD $1.9 billion in losses. Australian phone users sit firmly in the campaign window.

Disclosure: This post contains affiliate links. We only recommend tools we've researched and trust. If you purchase through our links, we may earn a commission at no extra cost to you.

Google's Lawsuit and What "Outsider Enterprise" Did

On 12 June 2026, Google filed a civil lawsuit in a New York federal court against a Chinese cybercriminal network the company calls "Outsider Enterprise." According to The Hacker News, this is the first lawsuit in which Google has explicitly pursued a group for weaponising its own Gemini AI platform to conduct large-scale phishing campaigns against consumers.

The complaint alleges that between November 2025 and April 2026 alone, the network generated more than 1.59 million fraudulent URLs linked to its infrastructure — roughly 9,000 of which were fully built-out fake websites. During just two weeks in May 2026, Android users filed 55,000 spam text complaints to Google, a rate of more than two per minute, while the gang sent 2.5 million phishing SMS messages containing links to those fraudulent sites.

The financial damage is staggering. The FBI's estimate cited in Google's complaint puts the total at approximately 3.87 million stolen credit card records and around USD $1.9 billion in losses traceable to Outsider Enterprise activity since July 2023, according to Help Net Security. Victims range from individual consumers who received a convincing fake toll payment text, to businesses whose employees clicked a plausible corporate IT login page generated by the service.

Outsider Enterprise operates as a phishing-as-a-service (PhaaS) platform. Criminal affiliates purchase access via Telegram channels and receive more than 290 pre-built phishing kit templates that convincingly impersonate financial institutions, phone carriers, government portals, toll payment systems, and retailers. One documented target was New York City's E-ZPass electronic toll system; another was the city government itself. Google, YouTube, and the US Postal Service were among the most-spoofed brands.

Google is seeking injunctive relief and damages. The complaint names multiple individuals and entities associated with the network, though the defendants are based in China and enforcement remains a separate challenge. The legal filing does, however, force public disclosure of the network's technical methods — and that detail is directly useful for defenders everywhere, including in Australia.

Why This Matters: PhaaS Has Industrialised Phishing for Any Criminal

The Outsider Enterprise lawsuit matters not because Google caught one criminal group, but because it exposes the industrial model now powering the majority of phishing campaigns globally. PhaaS platforms like Outsider Enterprise, Darcula, and the Group-IB-documented "Phoenix" system have dismantled the technical barrier that once separated sophisticated threat actors from opportunistic criminals. For under $1,000 a month — and sometimes far less — anyone with a Telegram account can rent a complete fraud operation, complete with AI-generated pages, backend panels, victim tracking dashboards, and live support.

Australia sits explicitly in the target window of these operations. Research from Palo Alto Networks' Unit 42 team confirms that Chinese-backed smishing services are designed to target victims simultaneously in the United States, the United Kingdom, Australia, and Japan within the same campaign window. The Outsider Enterprise infrastructure is consistent with that model: the 290 templates include localised variants for non-US markets, and the SMS delivery system uses carrier aggregators capable of routing messages through Australian mobile networks.

The scale also reframes how ordinary Australians should think about the SMS messages they receive. A message claiming your toll account has an outstanding balance, or that your parcel is held at customs, is no longer just an opportunistic attempt from one scammer. It may be generated by an AI, deployed by a criminal affiliate who has never written a line of code, using infrastructure maintained by an overseas organisation, all for a monthly subscription fee.

Google's own data from the complaint is sobering: phishing attacks generated using AI reportedly increased more than fourteenfold in late 2025 and now account for more than half of all reported phishing incidents. That figure comes from Google's own filing — the company is stating under oath the scale of AI-assisted fraud its own platform has been used to enable. It is a candid acknowledgement of a structural shift, not a vendor marketing claim.

For Australian small businesses, the second-order risk is equally concerning. A successful smishing attack against one employee can give an attacker authenticated access to SaaS platforms, cloud environments, accounting software, and internal communication tools — all without triggering a password-based alert if the victim was already logged in and the attacker harvests a session cookie rather than credentials directly.

How Gemini AI Was Weaponised: The Technical Detail

Google's complaint details the specific technique used to misuse Gemini. Outsider Enterprise operators were not breaking into Gemini or exploiting a vulnerability. They were simply prompt engineering around content policies by framing malicious requests as innocuous ones — for example, asking Gemini to build a "gift redemption" page or a "customer loyalty portal" that happened to look exactly like a major bank's login screen. The AI produced clean, professional HTML code that the operators then deployed on their phishing infrastructure.

The Prompt Engineering Angle

This approach — using AI coding assistants to generate phishing page HTML without triggering safety filters — is not unique to Gemini and has been documented across multiple major language models since late 2024. The attackers are exploiting the gap between what a model can detect as harmful (explicit requests for malware) and what it cannot easily detect (a request for a plausible-looking web form that happens to submit data to an attacker-controlled server). The end result is a phishing page that is harder to detect visually because it was generated by the same kind of AI used to build legitimate websites — consistent fonts, correct spacing, real logos sourced from public CDNs.

The Telegram Distribution Model

Outsider Enterprise sells access through Telegram channels. Affiliates purchase subscriptions, then browse a catalogue of 290 or more pre-built templates, selecting the brand they wish to impersonate and the territory they want to target. The platform handles hosting, SMS delivery, and victim data collection. Affiliates receive harvested credentials and card numbers through the same Telegram channel — they never need to maintain their own infrastructure.

This model is well-established across Chinese-language PhaaS platforms. The Group-IB Phoenix system documented earlier in 2026 operated the same way, including live-phishing interventions where a human operator joined an active phishing session to bypass real-time MFA prompts by relaying the victim's one-time code before it expired.

The Full Attack Chain Against an Australian Mobile User

Here is how the attack typically unfolds against an Australian target. The victim receives an SMS or RCS message appearing to come from a recognisable sender, such as a toll road operator or a bank. The message contains a shortened URL. Clicking it opens a convincing replica of the legitimate site, served over HTTPS with a plausible domain. The victim enters their card number, expiry, and CVV — or their banking login credentials. That data is immediately forwarded to the attacker's backend. The victim is then redirected to the legitimate site so the experience feels uninterrupted. Within hours, the stolen card data is listed on underground forums or sold in bulk to the Outsider Enterprise operators' own buyers.

In some campaigns, the attacker also requests a one-time SMS code, enabling them to bypass SMS-based two-factor authentication. The RCS messaging protocol used by Android devices actually assists this attack: because RCS shows message previews with link previews — including the fake site's branding — the recipient is more likely to interact before examining the URL carefully.

What Australians Can Do Right Now

There is no patch for phishing. Unlike a software vulnerability, this attack relies on human decision-making — which means the defences are behavioural, technical, and layered. Here is a practical checklist for Australian individuals and small business operators.

Never click a link in an unsolicited SMS or RCS message. If you receive a message claiming your toll account has an unpaid balance, your parcel is held, or your bank account needs verification, do not tap the link. Instead, open the official app directly or type the organisation's real URL into your browser manually. Legitimate organisations — including Services Australia, Australia Post, and every major Australian bank — do not require you to verify sensitive details via a link in a text message.

Report smishing attempts to the ACSC and Scamwatch. You can forward suspicious SMS messages to 7226 (SPAM), which feeds into the Australian Communications and Media Authority's complaint database. Reporting via Scamwatch also helps the ACCC track campaign volumes and issue public warnings. These reports actively shape what ACSC advisories are issued and what carriers are asked to block.

Use a dedicated password manager for every account — and make every password unique. The reason this matters in a smishing scenario: if you use the same password across multiple accounts and an attacker captures it via a phishing page, every account sharing that credential becomes vulnerable the moment the stolen data is tested. A password manager generates and stores a unique, long password for every site, so a single compromised credential does not cascade across your entire digital life.

NordPass is the password manager produced by the same security team behind NordVPN. It stores and autofills unique passwords, flags re-used credentials, and includes a data breach scanner that alerts you if any of your email addresses appears in a known credential dump — which is exactly what Outsider Enterprise's stolen card and login data ends up in. If you are still using the same password on more than one site, addressing that is the single highest-value security action you can take this weekend.

Enable authenticator-app-based 2FA, not SMS-based 2FA, wherever possible. The Outsider Enterprise campaign and similar PhaaS operations actively relay SMS one-time codes in real time to bypass SMS-based two-factor authentication. An authenticator app (Google Authenticator, Authy, or the authenticator built into NordPass) generates codes locally and cannot be intercepted in transit the same way an SMS code can. If your bank or a key supplier only offers SMS-based 2FA, that is better than nothing — but push for app-based or hardware token options where available.

Check your bank and card statements immediately. If you have clicked a link in an unexpected SMS in the past three months, check your accounts for unfamiliar transactions. Australian banks are generally fast to act on fraud reports; you can dispute charges within 60 days under the Australian Banking Code of Practice. The sooner you report, the higher your chance of a full reversal.

The Bigger Picture: AI Is Changing the Threat Landscape Permanently

Google's lawsuit is a legal landmark, but it is unlikely to significantly disrupt Outsider Enterprise's operations. The defendants are in China; US federal court judgements are largely unenforceable there. What the lawsuit does accomplish is creating a public court record of the technical methods, financial scale, and operational structure of a major PhaaS platform. That record is invaluable for security researchers, carriers, and platform providers who need to develop countermeasures — and it provides legal precedent for future actions against operators who misuse AI for fraud, regardless of where those operators are based.

The broader trend that the Google complaint describes — AI-generated phishing increasing fourteenfold — is the more urgent story. Outsider Enterprise is one named operator among many. The Darcula PhaaS platform, documented extensively by security researchers in 2025, uses the same Telegram distribution model and similarly targets Australian mobile users. The Phoenix PhaaS system uncovered by Group-IB adds real-time human operators who intervene during live phishing sessions to relay MFA codes. These platforms are evolving faster than the defences designed to catch them.

For Australian businesses, the implication is straightforward: your employees' phones are now as much a threat surface as their laptops. Security awareness training that covers "don't open suspicious email attachments" needs to be updated to include "don't tap links in unexpected SMS or RCS messages — even if the message appears to come from your bank, your employer's IT department, or a government agency." The sophistication of AI-generated phishing pages means that visual inspection of the resulting website is no longer a reliable defence. Checking the URL carefully before entering any credentials remains the single most reliable mitigation.

At the organisational level, Australian SMBs should consider reviewing what their employees can do with business credentials received via mobile. If a staff member can approve a payment, access a client database, or reset another user's password using credentials they entered on a mobile device, a successful smishing attack against that employee can translate directly into a business incident. Multi-factor authentication using an authenticator app — not SMS — and phishing-resistant options such as passkeys where available, are the architectural controls that reduce that exposure.

The ACSC's Annual Cyber Threat Report consistently identifies phishing as the most common initial access vector in Australian incidents. The arrival of AI-assisted PhaaS platforms that can target Australian phone users at scale, at low cost, using convincing localised content, makes that threat materially worse. Google's lawsuit is a signal that the major platforms are taking this seriously. The question for Australian individuals and businesses is whether their own defences have kept pace.

Practical summary for this week: verify every unexpected payment or account message through the official app or website, not the link in the message; ensure every important account uses a unique password stored in a manager; and switch SMS-based 2FA to an authenticator app wherever your accounts allow it. Those three steps address the specific attack chain Outsider Enterprise and similar operations rely on — and they are achievable over a weekend without specialist knowledge.