# MetaChat Brand Impersonation Phishing Campaign Targets AI API Keys and Credentials via EdgeOne Pages

> PreCrime Labs (BforeAI) uncovered an organized phishing operation impersonating MetaChat, an AI aggregation platform offering 30+ AI models (ChatGPT, Gemini, Claude, Midjourney, Grok, DeepSeek). Roughly 18 lookalike domains hosted on Tencent's free EdgeOne Pages platform harvest credentials, AI API keys/access tokens, payment data (including WeChat Pay), and referral network data via dedicated credential-harvesting endpoints and malvertising.

- **Published:** 2026-07-19T00:00:00Z
- **Last reviewed:** 2026-07-19T00:00:00Z
- **Canonical:** https://intel.threadlinqs.com/threat/TL-2026-1522
- **Card image:** https://intel.threadlinqs.com/og/threat/threat-intel-metachat-impersonation-edgeone-phishing-tl-2026-1522.webp?v=7b7a7e16 (Threadlinqs threat intel: MetaChat Phishing on EdgeOne Pages: Detection & IOCs (TL-2026-1522) — High severity, 34 IOCs, 9 detection rules.)
- **ID:** TL-2026-1522
- **Severity:** HIGH
- **Category:** PHISHING
- **Status:** ACTIVE
- **Detections:** 9 (1 shown free below) · **IOCs:** 34 (3 shown free below)

## Description

PreCrime Labs, the research division of BforeAI, identified a semi-automated phishing cluster impersonating MetaChat, a legitimate AI aggregation service that provides unified access to more than 30 AI models under one account, including ChatGPT, Gemini, Claude, Nano Banana, Midjourney, Grok, and DeepSeek. Approximately 18 lookalike domains were discovered, all hosted on Tencent's EdgeOne Pages free static-site hosting platform and following a predictable three-part naming convention: metachat-backend-[unique-id].edgeone.app, where [unique-id] is a randomly generated 10-character alphanumeric string. This naming pattern indicates semi-automated or scripted domain registration/provisioning rather than manual, one-off phishing kit deployment, enabling the operators to rapidly rotate infrastructure and outpace naive blocklist-based defenses.

Each phishing site clones the MetaChat login/registration flow and wires a set of purpose-built backend endpoints to harvest distinct categories of victim data: /user-register and /user-login capture account credentials (email, username, password, phone number); /user-api-key and /user-api-info capture AI API keys, access tokens, and account metadata for the platforms MetaChat aggregates; /user-recharge, /orders, and /chat/payment capture payment card and WeChat Pay transaction data tied to fraudulent 'recharge' (top-up) flows; /approval-token-withdraw captures internal session/approval tokens tied to withdrawal authorization; and /user-referees together with a SubAgentRefereeRechargePackage endpoint map the victim's referral network, indicating the operators are also running a two-tier MLM-style referral/commission fraud scheme layered on top of the credential theft.

Monetization is multi-pronged: stolen AI API keys and access tokens (for Claude, ChatGPT/Cursor, and other premium AI subscriptions) are resold on underground/cybercrime forums for unauthorized service consumption; WeChat Pay 'recharge' flows are abused to directly defraud victims of funds; and the referral/MLM structure incentivizes victims (and complicit distributors) to recruit further victims, amplifying reach. Because EdgeOne Pages is a free, low-friction hosting service backed by Tencent's legitimate CDN/edge infrastructure, phishing pages inherit a degree of implicit trust and evade reputation-based blocking that would flag cheaper or newly-registered standalone domains.

A secondary distribution vector uses malvertising via sponsored/paid search placements to route victims to lookalike pages impersonating other AI brands — for example, claude-app-new.gitlab.io impersonating Claude Code — placed above official links to exploit user purchase/download intent. This is consistent with a broader wave of AI-brand malvertising/impersonation activity observed across GitLab Pages and Google Ads in 2026 (a related, larger campaign used 92+ unique malicious GitLab Pages hostnames impersonating Claude Code, ChatGPT Codex, Perplexity, Cursor IDE, and JetBrains, paired with ClickFix-style social engineering to get victims to paste and execute malicious terminal/PowerShell commands). While the GitLab/ClickFix cluster is a distinct, malware-delivery-focused campaign, PreCrime Labs flags it as part of the same overall trend of attackers weaponizing the trust and hype around AI tools to target both credentials and endpoints.

Detection guidance from the source emphasizes flagging any newly registered subdomain on free hosting providers (EdgeOne Pages, GitLab Pages, and similar) that combines a 'backend'-style identifier with AI-specific keywords ('metachat', 'ai', 'api', 'chat', 'gpt', 'key') or financial-fraud keywords ('recharge', 'wallet', 'login'), particularly when the domain is registered/provisioned 30-60 days before becoming active in phishing distribution — a pattern PreCrime Labs' predictive models used to identify the cluster before full-scale abuse.

## MITRE ATT&CK

- T1583 Acquire Infrastructure
- T1584 Compromise Infrastructure
- T1587 Develop Capabilities
- T1585 Establish Accounts
- T1566 Phishing
- T1189 Drive-by Compromise
- T1204 User Execution
- T1528 Steal Application Access Token
- T1539 Steal Web Session Cookie
- T1056 Input Capture
- T1111 Multi-Factor Authentication Interception
- T1087 Account Discovery
- T1213 Data from Information Repositories
- T1119 Automated Collection
- T1102 Web Service
- T1090 Proxy
- T1567 Exfiltration Over Web Service
- T1657 Financial Theft
- T1531 Account Access Removal
- T1036 Masquerading
- T1090 Proxy

## Free detection (Splunk SPL, Microsoft KQL, Sigma)

**DNS/Proxy Resolution of MetaChat/Claude Lookalike Domain on Free Hosting** (`TL-2026-1522-SPL-001`)

Detects DNS or proxy resolution of hostnames matching the metachat-backend-[10-char-alphanumeric].edgeone.app naming convention, or claude*/claude-code*.gitlab.io lookalike domains impersonating AI brands, indicating a user has landed on the phishing initial-access page.

- **Languages:** Splunk SPL, Microsoft KQL, Sigma
- **ATT&CK:** T1583.001, T1566, T1566.002, T1036.005
- **Data sources:** network, proxy, dns
- **Severity:** high
- **Confidence:** high

### Splunk SPL

```spl
index=network sourcetype IN ("stream:dns", "pan:dns", "proxy") (query="*.edgeone.app" OR dest_host="*.edgeone.app" OR url="*.edgeone.app*")
| eval label=lower(coalesce(query, dest_host, url))
| regex label="metachat-backend-[a-z0-9]{10}\.edgeone\.app"
| eval matched_lookalike=if(match(label, "metachat-backend-[a-z0-9]{10}\.edgeone\.app") OR match(label, "claude(-code)?-app.*\.gitlab\.io"), "true", "false")
| search matched_lookalike=true
| stats count min(_time) as first_seen max(_time) as last_seen values(src_ip) as src_ips by label
| where count > 0
```

### Microsoft KQL

```kql
let AIBrandLookalikes = dynamic(["edgeone.app", "gitlab.io"]);
DnsEvents
| where TimeGenerated > ago(24h)
| where Name has_any (AIBrandLookalikes)
| where Name matches regex @"metachat-backend-[a-z0-9]{10}\.edgeone\.app" or Name matches regex @"claude(-code)?-app.*\.gitlab\.io"
| project TimeGenerated, Computer, ClientIP, Name, QueryType
| summarize FirstSeen=min(TimeGenerated), LastSeen=max(TimeGenerated), Count=count() by Name, Computer, ClientIP
```

### Sigma

```yaml
title: DNS Query for MetaChat/Claude AI-Brand Lookalike Domain
id: 1f6a2e10-1522-4a10-9a01-tl15220spl001
status: experimental
description: Detects DNS resolution of phishing domains matching the metachat-backend-[10-char-id].edgeone.app pattern or claude*.gitlab.io lookalikes used in AI-brand credential-harvesting phishing.
references:
    - https://bfore.ai/flash-notice/metachat-phishing-ai-api-key-theft/
author: Threadlinqs Team
date: 2026-07-19
tags:
    - attack.resource-development
    - attack.t1583.001
    - attack.defense-evasion
    - attack.t1036.005
    - attack.initial-access
    - attack.t1566.002
logsource:
    category: dns
detection:
    dga_pattern:
        query|re: 'metachat-backend-[a-z0-9]{10}\.edgeone\.app'
    gitlab_lure:
        query|contains|all:
            - 'claude'
            - '.gitlab.io'
    condition: dga_pattern or gitlab_lure
falsepositives:
    - None expected; the DGA-style regex and brand+gitlab.io combination are not used by legitimate services
level: high
```

**Known false positives:**

- Security researchers or the harness itself resolving these IOCs during threat-intel enrichment/sandboxing

**Tuning notes:** Maintain the regex against the current IOC feed as new metachat-backend-* domains rotate in; exclude known threat-intel sandbox/scanner source IPs to avoid analyst-tooling false positives.

### 8 more detection rules (Splunk SPL, Microsoft KQL, Sigma), unlock with TL Creds

- Rapid Credential-then-API-Key POST Sequence to Phishing Backend (locked: unlock with TL Creds)
- AI API-Key-Shaped Secret Posted to Non-Official Free-Hosting Domain (locked: unlock with TL Creds)
- Credential-Harvest Immediately Followed by Fraudulent Recharge/Payment POST (locked: unlock with TL Creds)
- Referral-Network Mapping POST to [indicator] Endpoint (locked: unlock with TL Creds)
- Fraudulent Withdrawal/Approval-Token POST to [indicator] (locked: unlock with TL Creds)
- Multi-Endpoint Exfiltration Chain Over Web Service to Trusted Free-Hosting CDN (locked: unlock with TL Creds)
- Sponsored-Search Referred Traffic Landing on AI-Brand Lookalike Page (locked: unlock with TL Creds)
- Newly-Registered Free-Hosting Subdomain Combining AI-Brand and Fraud Keywords (locked: unlock with TL Creds)

Sign up free to unlock the detection rules: https://threadlinqs.com/?modal=signup

## Sample indicators (3 of 34)

Defanged, TLP:CLEAR.

- Domain: `metachat-backend-if62u290lb[.]edgeone[.]app`
- Domain: `metachat-backend-yn4cnu2etz[.]edgeone[.]app`
- Domain: `metachat-backend-5gn5au3y14[.]edgeone[.]app`

Full set of 34 indicators unlocks with TL Creds: https://threadlinqs.com/tl-creds

## Sources

- [Scam Alert: MetaChat Impersonations Targeting AI API Keys](https://bfore.ai/flash-notice/metachat-phishing-ai-api-key-theft/)
- [EdgeOne Resource Abuse Prevention](https://www.tencentcloud.com/document/product/1145/63522)
- [Kaspersky has discovered a new corporate phishing technique using a popular AI web development platform](https://me-en.kaspersky.com/about/press-releases/kaspersky-has-discovered-a-new-corporate-phishing-technique-using-a-popular-ai-web-development-platform)
- [Kaspersky Flags AI Platform Abuse in Phishing Surge](https://techcoffeehouse.com/2026/06/15/kaspersky-flags-ai-platform-abuse-in-phishing-surge/)
- [Attackers abuse Google Ads, GitLab, and Claude to deliver malware](https://www.csoonline.com/article/4186813/attackers-abuse-google-ads-gitlab-and-claude-to-deliver-malware.html)
- [Fake Claude Code, Real Malware: Inside the Campaign Targeting AI Developers](https://www.straiker.ai/blog/acr-stealer-claude-code-impersonation-campaign)
- [Malicious Commands: Fake Claude Code & ChatGPT Installers](https://www.graphika.com/blogs/malicious-commands-fake-claude-code-chatgpt-installers)
- [Malware campaign impersonating Claude Code install via Google Ads](https://github.com/anthropics/claude-code/issues/34622)
- [AI-Assisted Phishing Campaign Harvesting Victim Data](https://cyble.com/blog/ai-assisted-phishing-campaign/)
- [AI-powered phishing campaign leverages browser permissions for data theft](https://www.scworld.com/brief/ai-powered-phishing-campaign-leverages-hardware-access-for-data-theft)

## Full data

The other 8 detection rules (Splunk SPL / Microsoft KQL / Sigma) unlock with TL Creds; a free account includes daily creds: https://threadlinqs.com/?modal=signup
Indicator values unlock with TL Creds, at a lower price on Red and Purple; AI agents can query rules and IOCs through the Threadlinqs MCP server (Purple): https://intel.threadlinqs.com/mcp

Canonical: https://intel.threadlinqs.com/threat/TL-2026-1522
