Open-source AI hacking agent AntiFox linked to live AWS, IIS breaches
Arctic Wolf Labs tied the AntiFox agentic pentest platform to confirmed intrusions into web apps and AWS environments
· Last verified: 13 Nov 2026 (Arctic Wolf Labs, Microsoft Threat Intelligence, AWS, Security Affairs, Techzine, Cloud Security Alliance)

Summary
- Arctic Wolf Labs tied the AntiFox agentic pentest platform to confirmed intrusions into web apps and AWS environments
- AntiFox chains reconnaissance, exploit validation and reporting across multiple AI agents with minimal human input
- Compromised app credentials and over-permissioned IAM roles are now the fastest route into enterprise cloud accounts
The latest
Attackers are running an open-source AI penetration-testing platform against live corporate targets, Arctic Wolf Labs reported on 10 November 2026, linking the tool known as AntiFox to confirmed compromises of internet-facing web services and the AWS environments behind them. Its consoles ran on Alibaba Cloud-hosted systems at two published source addresses, 47.79.4.33 and 8.208.101.210. No victims were named.
Details
- The platform: AntiFox describes itself as a self-driving AI penetration-testing platform built on a multi-agent design — a planner, workers and a lead agent handling reconnaissance, exploit validation and reporting. It is written in Go with an embedded Next.js interface and PostgreSQL storage, accepts Anthropic or OpenAI API keys, and ships under AGPL-3.0 with its console on TCP/8787.
- Rapid adoption: The repository was created on 9 October 2026, and its disclaimer restricts use to personal learning and local testing. GitHub stars and forks stood at 474 and 251 on 11 November, rising to 579 and 323 by 12 November, according to the Cloud Security Alliance — growth of roughly a fifth in a single day.
- The ARTEX link: The project profile describes AntiFox as a secondary development of the ARTEX platform from version v0.1.1. Microsoft Threat Intelligence reported on 5 November that attackers were using ARTEX to locate exposed secrets on GitHub and abuse them against cloud services. Arctic Wolf did not mention ARTEX.
- The IIS path: An internet-facing IIS application was breached through an unspecified vulnerability. The w3wp.exe process dropped ASPX web shells named pwn*.aspx and ran reconnaissance commands including whoami and icacls to map permissions and test write access to the site root directory. The intruder then used PowerShell and .NET configuration classes to read connection strings from web.config, probed an internal production database, and staged configuration files and logs in a web-accessible upload folder.
- The AWS path: A separate track began with valid AWS credentials of unknown origin, using IAM accounts and in some cases an assumed EC2 role. The actor probed identity and permissions through GetCallerIdentity, GetUser, ListUsers, ListRoles, ListPolicies and ListKeys, and attempted to obtain an ECR authorization token.
- Machine speed: In one intrusion the source pivoted across several AWS identities and issued hundreds of requests in about half an hour. Most returned AccessDenied, but some succeeded, and several chains ended in a successful GetSecretValue call against a production secret in Secrets Manager following ListSecrets.
- Data impact: In another case the attacker focused on S3, enumerating, reading, copying and deleting objects through GetObject, CopyObject and DeleteObject. Enumeration spanned IAM, EC2, S3, Lambda, Secrets Manager, KMS, RDS, ECS, EKS, Redshift, Athena, Glue, Systems Manager, API Gateway, CloudWatch, CloudFormation, DynamoDB and Amazon Bedrock. No Bedrock model invocation was reported.
- Amazon's position: AWS said its services "were not compromised and operated as designed," stating the activity relied on credentials taken from compromised customer applications. It recommended least privilege, enforcing IMDSv2 on EC2, keeping long-lived access keys out of application configuration, and enabling Amazon GuardDuty.
- The detection warning: Arctic Wolf cautioned defenders not to treat a request chain as failed simply because most calls were denied. Neither Arctic Wolf nor the independent reports named any affected organizations or attributed the activity to a specific threat actor, describing the intrusions only as multiple.
Background
The console title appeared in Chinese, translating as autonomous penetration-testing console. Security Affairs and Techzine reported the activity on 11 November, followed by a Cloud Security Alliance research note on 12 November.
Between the lines
The two intrusion paths converge on the same weakness: credentials stored where applications can read them. The IIS track ended at web.config connection strings; the AWS track began with credentials AWS says were pulled from compromised customer applications. The hundreds of requests in half an hour point to automation rather than manual operation, which is why high denial rates no longer signal a failed attempt.
What's next
Watch whether Arctic Wolf or Microsoft formally link AntiFox to ARTEX activity, whether GitHub acts against the repository, and whether its star and fork counts keep climbing at the pace recorded on 12 November.