Skip to navigation

Threat reportMalwareTL-2026-2686

x47.c Windows Botnet-as-a-Service Weaponizes xAI Grok for AI-Assisted Persistence and AI API Credit Draining

highACTIVE

A previously undocumented Windows botnet-as-a-service called x47.c, sold by the threat actor WraithTools on criminal marketplaces since early August 2026 (base $200, DDoS add-on $150, full package $950), bundles 18 DDoS methods, browser/Discord/wallet credential theft, SOCKS5 proxying, fast-flux C2, and a rootkit with two AI-specific innovations documented by Qrator Research Labs: an 'AI stealth' persistence module that queries the xAI Grok API to choose maintenance actions, and an 'AI API drain' command that uses stolen OpenAI/xAI-compatible API keys to run …

Key facts

IdentifiersTL-2026-2686
SeverityHIGH
Affected productsN/A Microsoft Windows (endpoint OS); OpenAI OpenAI API (chat/completions); xAI Grok API
AttributionWraithTools (low confidence)
Malwarex47.c, x47.c information stealer module
Toolsx47.c C2 panel
MITRE ATT&CK / ATLAS techniquesAML.T0034 Cost Harvesting, T1014 Rootkit, T1053.005 Scheduled Task, T1055.012 Process Hollowing, T1090.002 External Proxy, T1498.001 Direct Network Flood, T1498.002 Reflection Amplification, T1499.002 Endpoint Denial of Service, T1499.003 Application Exhaustion Flood, T1528 Steal Application Access Token, T1539 Steal Web Session Cookie, T1547.001 Registry Run Keys / Startup Folder and 5 more in the attack flow
MotivationFINANCIAL
Detection rules9 (1 shown free)
Sources9 cited sources

Detection coverage

As of 2026-09-29, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-2686 across Splunk SPL, Microsoft KQL and Sigma, covering 16 indicator(s) of compromise. The whole corpus is readable without an account. Every threat comes with one complete detection in Splunk SPL, Microsoft KQL and Sigma, free for everyone; a free account adds daily TL Creds to unlock the rest. Paid plans add a larger daily allowance, correlation and the MCP server. Threadlinqs MCP server · View plans.

Free detection: x47.c C2: SOCKS5 Proxy Relay, Fast-Flux DNS Rotation, and AI API Drain (Denial-of-Wallet) Traffic

Detects three C2/impact behaviors documented for x47.c: (1) SOCKS5 proxy relay traffic on ports 1080/9050 turning the host into a traffic relay for the botnet, (2) a fast-flux DNS resolution pattern (>=6 distinct domains and >=8 distinct resolved IPs per host in a 10-minute window) matching the panel's fast-flux configuration tab (6 domains/8 IPs per config), and (3) an abnormal burst of outbound calls to api.openai.com or api.x.ai (>100 calls in 5 minutes) consistent with the operator-issued 'AI API drain' command running up a victim's AI billing via a stolen key (denial-of-wallet / financial theft).

Splunk SPL

(index=network OR index=proxy OR index=firewall dest_port IN (1080,9050))
| eval marker="socks5_proxy_relay"
| append
    [ search index=dns sourcetype="dns"
      | bin _time span=10m
      | stats dc(query) as distinct_domains, dc(answer) as distinct_ips by src_ip, _time
      | where distinct_domains >= 6 AND distinct_ips >= 8
      | eval marker="fast_flux_resolution" ]
| append
    [ search (index=network OR index=proxy) dest_host IN ("api.openai.com","api.x.ai")
      | bin _time span=5m
      | stats count as api_calls by src_ip, _time
      | where api_calls > 100
      | eval marker="ai_api_drain_burst" ]
| stats values(marker) as behaviors, values(_time) as observed_times by src_ip
| table src_ip, behaviors, observed_times

Microsoft KQL

let socks5 = DeviceNetworkEvents
| where Timestamp > ago(24h)
| where RemotePort in (1080, 9050) and Protocol == "Tcp"
| summarize ConnCount = count() by DeviceName, RemoteIP, bin(Timestamp, 15m)
| where ConnCount > 20
| project DeviceName, Timestamp, Detail = strcat("SOCKS5 relay to ", RemoteIP, " (", ConnCount, " conns)");
let fastFlux = DeviceNetworkEvents
| where Timestamp > ago(24h)
| summarize DistinctDomains = dcount(RemoteUrl), DistinctIPs = dcount(RemoteIP) by DeviceName, bin(Timestamp, 10m)
| where DistinctDomains >= 6 and DistinctIPs >= 8
| project DeviceName, Timestamp, Detail = strcat("Fast-flux pattern: ", DistinctDomains, " domains / ", DistinctIPs, " IPs");
let aiDrain = DeviceNetworkEvents
| where Timestamp > ago(24h)
| where RemoteUrl has_any ("api.openai.com", "api.x.ai")
| summarize ApiCalls = count() by DeviceName, bin(Timestamp, 5m)
| where ApiCalls > 100
| project DeviceName, Timestamp, Detail = strcat("AI API drain burst: ", ApiCalls, " calls/5m");
union socks5, fastFlux, aiDrain
| order by Timestamp desc

Sigma

title: High-Volume Outbound Calls to OpenAI/xAI API Endpoints Consistent with AI API Drain (Denial-of-Wallet)
id: a5d9e3c1-4f27-4b6a-8e2d-1c9f6a3b7e04
status: experimental
description: Detects an abnormally high volume of outbound HTTPS requests to api.openai.com or api.x.ai from a single host/process within a short window, consistent with x47.c's operator-issued 'AI API drain' command, which uses a stolen OpenAI/xAI-compatible API key to run repeated billable requests directly against the provider (denial-of-wallet / OWASP LLM10:2025 Unbounded Consumption / MITRE ATLAS Cost Harvesting AML.T0034).
references:
  - https://qrator.net/blog/details/x47.c-botnet
  - https://cybernews.com/security/hackers-advertise-ai-token-draining-service/
author: Threadlinqs Team
date: 2026-09-27
tags:
  - attack.command_and_control
  - attack.impact
  - attack.t1090.002
  - attack.t1657
  - attack.tool.x47_c
logsource:
  category: network_connection
  product: windows
detection:
  selection:
    DestinationHostname|contains:
      - 'api.openai.com'
      - 'api.x.ai'
  filter_known_clients:
    Image|endswith:
      - '\chrome.exe'
      - '\msedge.exe'
      - '\firefox.exe'
  timeframe: 5m
  condition: (selection and not filter_known_clients) | count() by Image > 100
falsepositives:
  - Legitimate high-throughput AI application backends or batch-processing jobs that call the same provider APIs at volume
  - Load-testing or QA automation against AI provider sandbox/staging keys
level: high

Known false positives:

  • Legitimate SOCKS5 VPN/proxy clients used by remote workers on ports 1080/9050
  • High-throughput internal AI application services or batch jobs calling the same provider endpoints at volume
  • Enterprise DNS resolvers/CDNs with naturally high distinct-domain resolution counts

Tuning notes: Baseline per-host AI API call volume over 30 days before alerting on the 100-calls/5m threshold; exclude corporate egress/VPN concentrators from the fast-flux distinct-domain/IP heuristic; correlate AI API drain bursts with billing-anomaly alerts from the OpenAI/xAI provider dashboards for high-confidence triage.

8 more detection rules (Splunk SPL, Microsoft KQL, Sigma) · unlock with TL Creds. Sign up free to unlock the detection rules or query them with the Threadlinqs MCP server (Purple)
  • x47.c Stealer Module: Non-Browser Process Accessing Browser Credential Stores, Discord Tokens, or AI API Key Files · unlock with TL Creds
  • x47.c AI-Stealth Module: Persistence/Defense-Evasion Action Following Outbound Call to xAI Grok API · unlock with TL Creds
  • x47.c C2: SOCKS5 Proxy Relay, Fast-Flux DNS Rotation, and AI API Drain (Denial-of-Wallet) Traffic · unlock with TL Creds
  • x47.c Stealer Module: Non-Browser Process Accessing Browser Credential Stores, Discord Tokens, or AI API Key Files · unlock with TL Creds
  • x47.c AI-Stealth Module: Persistence/Defense-Evasion Action Following Outbound Call to xAI Grok API · unlock with TL Creds
  • x47.c C2: High-Volume Outbound Calls to OpenAI/xAI API Endpoints Consistent with AI API Drain (Denial-of-Wallet) · unlock with TL Creds
  • x47.c Stealer Module: Non-Browser Process Accessing Browser Credential Stores, Discord Tokens, or AI API Key Files · unlock with TL Creds
  • x47.c AI-Stealth Module: Persistence/Defense-Evasion Action Following Outbound Call to xAI Grok API · unlock with TL Creds

Indicators of compromise

As of 2026-09-29, Threadlinqs Intelligence tracks 16 indicators of compromise for TL-2026-2686.

16 indicators of compromise · unlock with TL Creds. Compare plans

Technical analysis

A previously undocumented Windows botnet-as-a-service called x47.c, sold by the threat actor WraithTools on criminal marketplaces since early August 2026 (base $200, DDoS add-on $150, full package $950), bundles 18 DDoS methods, browser/Discord/wallet credential theft, SOCKS5 proxying, fast-flux C2, and a rootkit with two AI-specific innovations documented by Qrator Research Labs: an 'AI stealth' persistence module that queries the xAI Grok API to choose maintenance actions, and an 'AI API drain' command that uses stolen OpenAI/xAI-compatible API keys to run up victims' AI billing in a denial-of-wallet attack.

Qrator Research Labs identified x47.c, a previously undocumented Windows botnet advertised by the threat actor WraithTools since early August 2026 (2026-08-03) on criminal marketplaces. The offering is tiered: a $200 base build, a $150 DDoS add-on, and a $950 complete package that includes credential theft, SOCKS5 proxying, and the AI-assisted persistence module. Qrator's analysis was based on the seller's advertisement, technical documentation, panel screenshots, and follow-up messages from WraithTools, rather than a captured live sample.

The operator-facing command-and-control panel exposes bot management, a fast-flux configuration tab listing 6 rotating domains and 8 IP addresses per configuration, information-stealer log access, proxy health/timeout monitoring, and a DDoS tab offering 18 attack methods (HTTP floods, slow HTTP connections, TCP and UDP floods, a TLS stresser, and several reflection/amplification techniques, alongside the AI API drain method described below). An information-stealer component harvests browser-stored passwords and cookies, Discord authentication tokens, cryptocurrency wallet data, and AI-service (OpenAI/xAI) API tokens. A SOCKS5 proxy module converts infected hosts into traffic relays, and an optional rootkit module removes competing malware to preserve exclusive control of the host.

The 'AI stealth' module is x47.c's headline persistence innovation: it calls the xAI Grok API, embedding an xAI API key in the build, to have the model assess the infected host and select from a predefined menu of maintenance/concealment actions — creating startup (Run key) entries, registering scheduled tasks, repairing broken persistence, and configuring Windows Defender exclusions. The module falls back to local, hardcoded logic if the API call fails, and reports status messages back on the actions taken. Optional process-hollowing and privilege-escalation features are gated on the presence of a valid xAI API key in the build.

The 'AI API drain' command is a distinct, billing-focused abuse primitive: the operator supplies a target model name and a valid (typically stolen) API key for OpenAI, xAI, or a compatible chat-completions API, and the botnet issues repeated billable requests directly to the provider. Because the requests go straight to the provider rather than through the victim's own application, the victim's site or product can remain fully available while its AI usage bill or prepaid credit balance is exhausted — the pattern OWASP's GenAI security guidance classifies as Denial of Wallet (DoW) under LLM10:2025 Unbounded Consumption. Filtering or rate-limiting at the victim's application layer does not stop the charges, since the abusive traffic never traverses it. Qrator and follow-on reporting recommend revoking/rotating exposed AI API keys, monitoring billing for anomalous request volume, enabling spending caps, disabling provider auto-recharge, endpoint cleanup for infected hosts, and layered DDoS mitigation for the traditional flood methods. No CVE, CVSS score, or specific network/file IOCs (domains, IPs, hashes) have been publicly disclosed for x47.c as of this writing.

MITRE ATT&CK / ATLAS techniques by tactic:

Impact

AML.T0034 Cost Harvesting; T1498.001 Direct Network Flood; T1498.002 Reflection Amplification; T1499.002 Endpoint Denial of Service; T1499.003 Application Exhaustion Flood; T1657 Financial Theft

Defense Evasion

T1014 Rootkit

Persistence

T1053.005 Scheduled Task; T1547.001 Registry Run Keys / Startup Folder

Privilege Escalation

T1055.012 Process Hollowing

Command and Control

T1090.002 External Proxy; T1568.001 Fast Flux DNS

Credential Access

T1528 Steal Application Access Token; T1539 Steal Web Session Cookie; T1555.003 Credentials from Web Browsers

Resource Development

T1583.001 Domains

defense-impairment

T1685 Disable or Modify Tools

Remediation

Immediate actions

  • Revoke and rotate any OpenAI, xAI, or compatible chat-API keys suspected of exposure or embedding in third-party/unofficial software
  • Audit AI-provider billing and usage logs for anomalous request volume or spend inconsistent with legitimate application traffic
  • Enable spending caps / hard billing limits and disable auto-recharge on AI provider accounts to bound denial-of-wallet exposure

Workarounds

  • Disable AI-provider auto-recharge and enforce per-key rate limits to cap denial-of-wallet exposure even if a key is stolen
  • Restrict/monitor outbound SOCKS5 and non-standard proxy traffic from endpoints to curb proxy-relay abuse
  • Perform endpoint cleanup or reimaging on hosts suspected of x47.c infection given the bundled rootkit and persistence-repair capability

Longer-term hardening

  • Deploy layered, multi-vector DDoS mitigation capable of absorbing HTTP floods, slow HTTP, TCP/UDP floods, TLS stress, and reflection/amplification traffic
  • Harden endpoint credential exposure (browser-stored passwords/cookies, Discord tokens, wallet files) against commodity stealer modules
  • Monitor for unauthorized startup-entry and scheduled-task creation, and for Windows Defender exclusion changes, as persistence indicators

Affected and fixed versions

  • N/A — Microsoft Windows (endpoint OS)
    Vulnerable versions: Not version-specific — x47.c is a Windows-targeting commodity botnet-as-a-service, not a vulnerability in a specific Windows release
  • OpenAI — OpenAI API (chat/completions)
    Vulnerable versions: Any account whose API key is compromised or stolen
  • xAI — Grok API
    Vulnerable versions: Any account whose API key is compromised or stolen

Timeline

Last updated · revalidated 1× · latest source

  • WraithTools begins advertising the x47.c Windows botnet-as-a-service on criminal marketplaces: $200 base package, $150 DDoS add-on, $950 full package bundling credential theft, SOCKS5 proxying, and the AI-stealth persistence module.
  • Qrator reports no confirmed infection count, measured attack capacity, verified victim losses, or tests of the advertised protection bypasses; the capabilities are seller-advertised and documented rather than observed from a captured sample.
  • Cybernews and PrivacyNeedle publish early coverage of Qrator's findings, framing the AI API drain feature as a denial-of-wallet (DoW) attack and detailing recommended defenses (key rotation, spending caps, disabling auto-recharge).
  • Qrator Research Labs publishes its technical analysis of x47.c, based on the seller's advertisement, technical documentation, C2 panel screenshots, and follow-up seller messages.
  • Threadlinqs threat-intel pipeline ingests the SecurityWeek article via the monitored SecurityWeek RSS feed and opens threat TL-2026-2686 for research.
  • Infosecurity Magazine, SC Media, CloudLinkTech, and News4Hackers publish follow-on coverage detailing the 18 DDoS attack methods, the fast-flux C2 configuration (6 domains / 8 IPs per tab), and the AI-stealth persistence mechanics.
  • SecurityWeek publishes "New x47.c Windows Botnet Weaponizes xAI Grok, AI API Draining," bringing the Qrator findings to mainstream security media; SecurityWeek is the article ingested to open this threat.
  • Additional outlets (e.g., HendryAdrian) continue to syndicate the x47.c findings, cross-referencing MITRE ATT&CK-relevant behaviors such as browser credential theft and lateral infrastructure abuse.
  • Cybersecurity News reports the first concrete artifact names for x47.c: locked-ioc-2, locked-ioc-3, locked-ioc-1, package directory x47.c_FF_v4.1 and the relay handshake prefix 'REVERSE_PROXY|'; domain and IP values remain withheld.

Update history

Sources cited

Previous and next in malware

Drafted by the Threadlinqs research pipeline from 9 sources; published ; last updated .

Threadlinqs Intelligence — Real-Time Threat Detection Platform

// threat_feed
$ sort --newest
Showing all threats

Live intelligence console

Threat level

Latest Threats