> ## Content Index
> Fetch the complete content index at: https://f4n6.co.uk/llms.txt
> Use this file to discover other available public pages before exploring further.

# The fake sites using a cheap toolkit to sell $2,000 AI subscriptions
- URL: https://f4n6.co.uk/security-feed/the-fake-sites-using-a-cheap-toolkit-to-sell-2-000-ai-subscriptions/
- Published: 2026-09-21T21:37:17.000Z
- Updated: 2026-09-21T21:37:17.000Z
- Author: Jeff Davies
- Tags: #security-feed

## 1\. Executive summary

Malwarebytes identified a network of more than 100 subscription websites built on the same commercial website starter kit and linked through closely related Google developer contact addresses, consistent with a single operator or a closely connected group. The sites impersonate real products including GPT-6 Astra, DaVinci Resolve, PixAI and OpenCut, and one reuses the name of Omegle, which shut down in 2023; others promote unfamiliar brands with no verifiable operator. The sites do not push malware or harvest passwords directly. They use polished product pages and genuine Google OAuth sign-in to sell subscriptions priced from under $10 a month to over $2,000 a year, and several ask users to upload documents, recordings or other files. For EMEA financial services the exposure is card-payment fraud, disputed charges and disclosure of client or firm material to an unidentifiable operator, not a technical compromise. No CVEs are involved and no named threat actor is identified, so there is no verified severity rating or attribution to report.

## 2\. Regulatory framing

No specific DORA/NIS2 article is directly engaged by this item. The item is a consumer-facing subscription-fraud network reported by a single vendor; it does not describe an ICT incident at a financial entity, a compromised third party, or a supply-chain dependency that a client must act on under the articles in scope.

## 3\. Technical analysis & attack chain

This is a fraud operation, not an intrusion. There is no malware, no exploited vulnerability and no credential-theft page in the observed chain. The mechanism is a subscription storefront network, and the attack chain runs against the buyer's trust rather than their device.

Confirmed steps, from the Malwarebytes report:

1. The operator buys a legitimate commercial website starter kit, priced at $249 as a one-time purchase with additional templates at about $2 each. The kit ships with an account system, billing, file storage and administrative tools; the buyer supplies only branding and a product description. The kit's vendor advertises that a buyer can launch a product in an hour.
2. The operator deploys more than 100 sites from the kit. Each site presents a distinct brand: some impersonate existing products (GPT-6 Astra, DaVinci Resolve, PixAI, OpenCut), one reuses the defunct Omegle name, and others promote unfamiliar brands. Advertised categories include voice-cloning tools, video editors, study apps and general-purpose AI assistants.
3. Landing pages carry fabricated social proof: performance comparisons, star ratings, precise user counts (one site claimed more than 12 million users) and well-known company names displayed as customers with no evidence of any connection. Several sites retain demonstration material shipped with the kit, including the kit's own brand name, promotional banners, generic menu entries and stock testimonials. One site relabelled the kit's demonstration business list as its own customers; another left the word "boilerplate" in the name of a paid subscription plan.
4. The visitor clicks Sign in or Get started and is routed to a genuine Google sign-in page. The URL belongs to Google and the password is entered on Google's site, not on the fraudulent service. The sites requested basic profile information: name, email address and profile picture. They did not request access to Gmail or Google Drive.
5. After sign-in the visitor is shown paid plans, usage credits and payment histories rather than the advertised tool. There is no way to test the product before paying; landing-page controls did nothing until sign-in, and sign-in led to a pricing page. Prices range from under $10 a month to over $2,000 a year, with annual plans charged as a single payment and some plans stating that unused credits expire after a set period. Several sites ask users to upload documents, recordings or other files for the advertised service to process.

How Malwarebytes linked the network. Two independent signals connect the sites. First, the sites load files whose names are derived from file contents, and many of these filenames were identical across the sites, indicating the same version of the same underlying software. On its own that would only show that different people bought the same kit. The stronger signal is the Google developer contact details, which are entered when an application is registered with Google rather than supplied automatically by the kit: the developer contacts are free webmail accounts containing the same name, differing only in small variations such as appended numbers. That points to one operator or closely connected people.

Operator opacity. The sites provide no registered company names, business addresses or other independently verifiable ownership details. In many cases the only contact method is an email address on the site's own domain. Depending on the payment system used, card details may be handled by a third-party payment provider rather than the site operator, but users still share account information and potentially sensitive uploaded material with an owner who cannot be identified.

Confidence caveat. Every finding above rests on a single vendor report (Malwarebytes, 2026-09-21). The site count, the kit linkage, the developer-email linkage and the impersonation claims are single-sourced; verify before enforcement. No domains, file hashes or specific webmail addresses were published in the source material, so no IOC set can be extracted from it. No threat actor is named, and no MITRE ATT&CK profile exists in the verified reference data, so attribution is unconfirmed.

## 4\. Mitigation & containment

This item implicates payment verification, expense-review and user-education controls rather than technical containment. There is no patch to apply and no CVE to track.

### P1, within 24 hours

- Search corporate card and expense records for charges matching the profile in the report: recurring subscriptions from under $10 a month, or single annual payments up to and above $2,000, from unfamiliar AI, video-editing, voice-cloning or study-app vendors. Flag any charge where the vendor cannot be matched to a registered company.
- Ask staff who used AI subscription tools in the last 90 days whether they signed in with Google on a site advertising GPT-6 Astra, DaVinci Resolve, PixAI, OpenCut or Omegle-branded services, or on any AI tool site with no verifiable operator. Where the answer is yes, treat any documents, recordings or files uploaded to that service as disclosed to an unidentified third party and assess them accordingly.
- Instruct staff to check the application name and developer contact shown on Google's OAuth consent screen against the brand on the site before signing in. A free webmail address as the developer contact for a service claiming millions of users is the report's clearest tell.

### P2, within 72 hours

- Add the impersonated brand names (GPT-6 Astra, DaVinci Resolve, PixAI, OpenCut, Omegle) to web-filtering and DNS monitoring watchlists as search-driven lures, pending publication of the specific domains by Malwarebytes or follow-on researchers. Do not block the genuine products' domains.
- Update anti-fraud and payment-verification procedures to cover subscription charges from vendors with no registered company name, no business address and only a domain-based contact email, which the report identifies as the network's common profile.
- Brief helpdesk and finance teams that a valid TLS padlock and a genuine Google sign-in page confirm encryption and Google's handling of the login, not that the requesting service is legitimate.

### P3, within 7 days

- Fold the report's verification checklist into staff security-awareness content: cross-check Google consent-screen developer details against the site's brand, treat on-page ratings and user counts as self-reported, and treat paywall-before-demo as a warning sign.
- Add a retrospective review of the past year's software-subscription spend for charges from vendors that cannot be verified against a company registration, and reclaim where possible.

## 5\. Indicators of compromise

No indicators of compromise available in the source material. Malwarebytes did not publish domains, IP addresses, file hashes, webmail addresses or kit filenames in the report.

Behavioural indicators:

| Behaviour                                                                                                                            | Where to observe                              | Confidence               |
| ------------------------------------------------------------------------------------------------------------------------------------ | --------------------------------------------- | ------------------------ |
| Subscription charge from an AI, video-editing, voice-cloning or study-app vendor with no registered company name or business address | Corporate card and expense-management records | Moderate, single-sourced |
| Google OAuth consent screen showing a developer contact at a free webmail address that does not match the displayed brand            | Browser, at the consent prompt                | Moderate, single-sourced |
| Landing page with no working product controls before sign-in, where sign-in leads to a pricing page rather than the tool             | Browser, on the site                          | Moderate, single-sourced |
| Paid subscription plan names or customer lists containing starter-kit demonstration material, including the word "boilerplate"       | Browser, on the site                          | Moderate, single-sourced |

## 6\. Detection

Insufficient indicators to author detection rules. The source material contains no domains, file hashes, distinctive strings, command-line artefacts or log events to build a YARA or Sigma rule from. The behavioural indicators in §5 are procedural and belong in fraud and awareness controls, not detection engineering.

## 7\. Sources

- Malwarebytes, "The fake sites using a cheap toolkit to sell $2,000 AI subscriptions", https://www.malwarebytes.com/blog/threat-intel/2026/09/the-fake-sites-using-a-cheap-toolkit-to-sell-2000-ai-subscriptions, 2026-09-21

## 8\. Adverse Trace position

We assess this as a low-severity but operationally relevant item for EMEA financial services: no intrusion, no malware and no CVE, but a demonstrated fraud pattern that reaches staff and corporate cards through search and polished branding, and that can move client-adjacent material to an unidentifiable operator through file-upload prompts. The entire dataset is single-sourced from Malwarebytes, and the operator attribution rests on kit-file linkage plus Google developer contact patterns rather than on a named actor; we treat both as unconfirmed pending corroboration. We will monitor for publication of the specific domains and webmail addresses, for follow-on reporting from other vendors, and for any evidence that uploaded documents were resold or leaked, and we will reissue this advisory with an IOC set if those details appear.

---

[Read the original source →](https://www.malwarebytes.com/blog/threat-intel/2026/09/the-fake-sites-using-a-cheap-toolkit-to-sell-2000-ai-subscriptions?ref=f4n6.co.uk)

*Published via PulseTrace — Adverse Trace threat intelligence.*