# AnonyMousKIT: AI-Powered Phishing-as-a-Service Platform Stealing Apple IDs from Stolen iPhones

> SOCRadar's Threat Research Unit (STRU) exposed AnonyMousKIT, an AI-powered Phishing-as-a-Service platform that turns Apple's Activation Lock into a subscription-based crime: operators load a stolen iPhone's real model number and live Find My location into a panel, then automated email, SMS, WhatsApp, recorded-call, and AI-voice ('Alice from Apple Support') lures run until the owner is tricked into surrendering their Apple ID and passcode. A backend misconfiguration leaked months of production logs, exposing a shared codebase running 506 domains under 168 brand names across a four-tier criminal supply chain (developer, resellers, 689 WhatsApp operator accounts, and paying buyers), including lures sent to 27 South African government email addresses and a local university.

- **Published:** 2026-08-24T00:00:00Z
- **Last reviewed:** 2026-08-24T00:00:00Z
- **Canonical:** https://intel.threadlinqs.com/threat/TL-2026-2141
- **Card image:** https://intel.threadlinqs.com/og/threat/threat-intel-anonymouskit-apple-id-theft-tl-2026-2141.webp?v=9b2e1372 (Threadlinqs threat intel: AnonyMousKIT AI Voice PhaaS: Detection & IOCs (TL-2026-2141) — High severity, 13 IOCs, 9 detection rules.)
- **ID:** TL-2026-2141
- **Severity:** HIGH
- **Category:** PHISHING
- **Status:** ACTIVE
- **Detections:** 9 (1 shown free below) · **IOCs:** 13

## Description

AnonyMousKIT is an AI-powered Phishing-as-a-Service (PhaaS) platform, disclosed by SOCRadar's Threat Research Unit (STRU) on 2026-08-24, that exists specifically to steal the Apple ID credential and device passcode a criminal needs to clear Activation Lock on a stolen iPhone and resell or trade it, including on marketplaces such as Telegram. STRU traces the underlying kit family back to at least February 2024, with active operations continuing through August 2026 (panel logs record operator activity as recently as 2026-08-10). The business model is credit-metered/pay-per-message: a buyer (typically a phone thief or a reseller downstream of one) enters the stolen device's details into a web panel once, and the platform then runs an automated, multi-channel harassment campaign against the device's original owner across email, SMS (smishing), WhatsApp, an automated recorded phone call, and an AI-generated voice phone call — continuing until the victim responds.

The lures are device-led: rather than generic phishing bait, each message cites the phone's real Apple model identifier (e.g., iPhone16,2) and its live Find My location, both pulled directly off the physically stolen device, which lends the lure false legitimacy. The phishing landing pages carry that device-led deception further with anti-bot checks, localized/multilingual content, and an interactive animated map that renders the handset's real-time Find My status and location to build trust and urgency before sequentially harvesting the device passcode, then the Apple ID credentials, then a live 2FA code — each captured value forwarded from the page to both the operator's panel and a Telegram webhook. Across 30 related backends, researchers identified 6,092 phishing emails sent targeting 5,031 devices marked online and 1,035 locked devices; those email backends run on 24 identified SMTP instances, 12 of which share an identical developer test pattern that STRU used to fingerprint a single core developer behind the platform.

The AI voice channel goes further, renting commercial conversational AI agents from VAPI.ai and scripting them as 'Alice from Apple Support,' delivered in English, Spanish, and Portuguese; 'Alice' confirms ownership of the stolen device, claims Apple recovered the missing phone, and talks the victim into reading their 4-6 digit passcode (or a 2FA code) aloud before redirecting them to a phishing link to capture their Apple ID credentials. SOCRadar recovered 200 call records and 55 transcripts — including the agents' underlying prompts — directly from the VAPI.ai account, with 179 of the 200 calls (roughly 90%) placed to Brazilian numbers. The economics make indiscriminate targeting viable: the operator's cost for those 200 AI voice calls was only $19.24 total, about ten cents per call.

STRU's disclosure resulted from a basic backend coding mistake: two exposed relative file paths, left accessible without authentication, handed researchers months of production logs, which traced a single storefront back to a shared codebase powering 506 domains under 168 distinct brand names — evidence of a reseller network rebranding the same kit. The leaked logs reveal a four-tier criminal supply chain beneath that storefront layer: a single developer (fingerprinted via the shared SMTP test pattern); a reseller tier of 506 domains / 168 brands built on the shared codebase; an operator tier of 689 distinct WhatsApp accounts running phishing operations across the 30-backend family; and a buyer tier of 27 paying email-based customers sourced from a pool of 120 WhatsApp users. A scan found 30 backend instances still active across 42 of those domains at the time of research.

While the kit's stated purpose is unlocking stolen consumer iPhones, STRU's research documents lures sent to 27 South African government email addresses and a local South African university — targeting that extends the risk beyond individual device theft victims. Because a personal Apple ID can be tied to a device's Keychain, a successful takeover of a government or university employee's personal Apple ID creates a plausible path to exposure of corporate credentials synced to that Keychain, an organizational risk distinct from the underlying stolen-phone fraud.

No CVE, CVSS score, or software vulnerability is associated with this threat — AnonyMousKIT is a social-engineering/PhaaS operation, not an exploit against a technical flaw in Apple's platform; the only technical weakness in the reporting is the operators' own exposed relative file paths, which is what enabled the disclosure.

## MITRE ATT&CK

- T1589 Gather Victim Identity Information
- T1598 Phishing for Information
- T1583 Acquire Infrastructure
- T1608 Stage Capabilities
- T1585 Establish Accounts
- T1566 Phishing
- T1684.001 Impersonation
- T1056 Input Capture
- T1567 Exfiltration Over Web Service
- T1657 Financial Theft

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

**Outbound Credential Exfiltration to Telegram Bot API Webhook Following Phishing Landing-Page Visit** (`TL-2026-2141-SPL-003`)

Detects an outbound HTTPS POST to the Telegram Bot API (api.telegram.org/bot<token>/...) from a managed asset within minutes of that same asset visiting an AnonyMousKIT-pattern phishing domain — the network signature of the landing page mirroring a captured device passcode, Apple ID credential, or live 2FA code out to the operator's Telegram webhook, in addition to the operator's own panel.

- **Languages:** Splunk SPL, Microsoft KQL, Sigma
- **ATT&CK:** T1567, T1608.005
- **Data sources:** ngfw_traffic_logs (Palo Alto/Cisco ASA), ngfw_proxy_logs, m365_device_network_events
- **Severity:** high
- **Confidence:** high

### Splunk SPL

```spl
index=firewall (sourcetype=pan:traffic OR sourcetype=cisco:asa)
| where dest_host="api.telegram.org" AND http_method="POST"
| eval telegram_time=_time
| join type=inner src_ip [ search index=proxy (sourcetype=paloalto:threat OR sourcetype=zscalernss-web) earliest=-15m
  | eval url_lc=lower(url)
  | where match(url_lc,"passcode") OR match(url_lc,"appleid") OR match(url_lc,"apple-id") OR match(url_lc,"2fa") OR match(url_lc,"otp")
  | table src_ip url_lc _time ]
| where (telegram_time - _time) <= 900 AND (telegram_time - _time) >= 0
| stats count min(_time) as phishing_visit_time min(telegram_time) as exfil_time values(url_lc) as phishing_urls by src_ip
| where count>0
```

### Microsoft KQL

```kql
let PhishingVisits = DeviceNetworkEvents
| where RemoteUrl has_any ("passcode","appleid","apple-id","2fa","otp")
| project DeviceId, PhishTime=Timestamp, PhishUrl=RemoteUrl;
DeviceNetworkEvents
| where RemoteUrl has "api.telegram.org"
| where Timestamp > ago(24h)
| join kind=inner PhishingVisits on DeviceId
| where Timestamp - PhishTime between (0min .. 15min)
| summarize ExfilEvents=count(), PhishUrls=make_set(PhishUrl) by DeviceId
| where ExfilEvents > 0
```

### Sigma

```yaml
title: Outbound POST to Telegram Bot API Shortly After Phishing Credential-Harvesting Page Visit
id: 6d3a8e1f-4c9b-4f2a-8e5d-1a7c9b3f6e04
status: experimental
description: Detects an outbound HTTPS POST to the Telegram Bot API (api.telegram.org) from a host, intended to be correlated (via SIEM join, within a 15-minute window) against that host having visited a credential-harvesting-pattern URL, consistent with AnonyMousKIT's captured-credential mirroring to a Telegram webhook.
references:
  - https://socradar.io/blog/anonymouskit-ai-phaas-supply-chain/
author: Threadlinqs Team
date: 2026-08-25
tags:
  - attack.exfiltration
  - attack.t1567
  - attack.resource-development
  - attack.t1608.005
logsource:
  category: proxy
detection:
  selection_telegram:
    c-uri|contains: 'api.telegram.org'
    cs-method: 'POST'
  condition: selection_telegram
falsepositives:
  - Legitimate business use of a Telegram bot integration (e.g., internal alerting/notification bot) unrelated to this threat
level: high
```

**Known false positives:**

- Legitimate internal Telegram bot integration (alerting, chat-ops) unrelated to phishing
- Shared corporate NAT egress IP causing an unrelated user's Telegram traffic to fall inside the 15-minute correlation window

**Tuning notes:** Maintain an allowlist of known-legitimate internal Telegram bot tokens/chat IDs; tighten or widen the 15-minute join window against observed AnonyMousKIT page-load-to-exfil timing once available.

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

- Safe Links Click-Through to AnonyMousKIT Credential-Harvesting Landing Page (locked: unlock with TL Creds)
- Multi-Parameter Sequential Credential POST to Single External Domain (Passcode → Apple ID → 2FA) (locked: unlock with TL Creds)
- Outbound Reuse of VAPI.ai Conversational-AI Infrastructure From Unexpected Corporate Assets (locked: unlock with TL Creds)
- First Contact With Newly Registered Apple/Find My Branded Domain (No Email Referrer) (locked: unlock with TL Creds)
- AnonyMousKIT Landing-Page Fingerprint (Anti-Bot Check + Animated Find My Map) (locked: unlock with TL Creds)
- Corporate-Managed Device iCloud Keychain Sync Contact Following AnonyMousKIT Phishing Exposure (locked: unlock with TL Creds)
- AnonyMousKIT Device-Led Apple/Find My Phishing Email — Initial Lure (locked: unlock with TL Creds)
- AI-Voice 'Apple Support' Call Correlated With Sequential Passcode/Apple ID/2FA Harvesting Session (locked: unlock with TL Creds)

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

## Sources

- [Exposing AnonyMousKIT: AI-Powered PhaaS Supply Chain](https://socradar.io/blog/anonymouskit-ai-phaas-supply-chain/)
- [SOCRadar Uncovers AI-Powered PhaaS "AnonyMousKIT" Stealing Apple IDs/Passwords](https://itnerd.blog/2026/08/24/socradar-uncovers-ai-powered-phaas-anonymouskit-stealing-apple-ids-passwords/)
- [AnonyMousKIT Campaign Details — SOCRadar LABS](https://socradar.io/labs/campaigns/1212/details)
- [AnonyMousKIT service uses AI calls to unlock stolen Apple devices](https://cyberinsider.com/anonymouskit-service-uses-ai-calls-to-unlock-stolen-apple-devices/)
- [Hackers using fake "Apple Support" calls to unlock stolen iPhones](https://cybernews.com/cybercrime/stolen-iphone-ai-phishing-apple-support-scam-passcodes/)
- [SOCRadar — "AI is now cold-calling iPhone theft victims" (AnonyMousKIT disclosure thread)](https://x.com/socradar/status/2091867195437629745)

## 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-2026-2141
