Iris C2Agent: hunting malware that hands control to an AI model
A look at a new LLM-operated Windows implant, what it leaves behind, and how to investigate it. By the Container 7 Team Most implants wait for an operator to send commands. Iris, also tracked as C2Agent, waits for a goal. The operator types something like "get the administrator's password" and an AI model takes it from there, choosing each command, reading the output and deciding what to do next. If you defend Windows endpoints in an organization that uses AI tools, this one's worth your time. What's happening Iris is a Windows implant written in Go. It was first seen on September 10, 2026, and Palo Alto Networks' Unit 42 documented it in October 2026. Its design is unusual. The operator doesn't send a script or a list of commands. They send a plain-language objective, and the implant hands interactive control of the host to a large language model (LLM). The model works in a loop. It runs a command, reads the result and picks the next step. The family has moved fast. Unit 42 tracked 32 samples across eight major versions in 29 days. Newer versions can hijack AI developer tools already installed on the victim's machine and abuse legitimate commercial AI services to carry out their work. Unit 42 also describes it as the first documented implant to act as a Model Context Protocol (MCP) client, which lets it add new tool connections at runtime. Why it matters Iris changes two things defenders rely on. First, there's no fixed playbook. A traditional implant repeats the same tasks in the same order, which gives you a pattern to detect. With a model in control, the sequence of commands can shift from one run to the next, even when the goal stays the same. Second, it hides in traffic you've probably already approved. Most organizations allow AI services out of the network because their developers and analysts use them every day. When the malware calls those same services, or drives an AI tool that's already installed, malicious requests sit next to legitimate ones. Your AI egress policy now affects how much you can see. There's also an upside for defenders. A model that reasons step by step can leave a record of that reasoning. If you can find it, you can see what the attacker asked for and what the model tried to do. What to look for No single signature covers every version, so hunt for behavior. Start with these questions: Process lineage: Are AI developer tools or command-line assistants being launched by unexpected parent processes, or running commands nobody on the team asked for? AI egress: Which hosts talk to commercial AI services, and do those hosts have a business reason to? Look for AI traffic from servers or accounts that don't normally use it. Command cadence: Look for bursts of reconnaissance and credential-access commands with short, regular gaps between them, as if each one waited for the last result. On-disk artifacts: Watch for new configuration, prompt or "skill" files appearing alongside unfamiliar binaries in user-writable folders. Recorded reasoning: Where you can find the model's output, compare what it planned with what the host logs say actually ran. Get hands-on The Iris C2Agent Log Analysis lab puts you in the defender's seat. Working in Elastic, you'll investigate a Windows host where Iris was executed and tasked by a remote operator. You'll recover the artifacts it left on disk and on the network, then read the model's own recorded reasoning to separate what the attacker intended from what actually happened on the host. By the end, you'll know the indicators of compromise (IOCs) this family leaves behind. The lab is built for security analysts, incident responders, threat hunters, and detection engineers who monitor Windows estates. It also suits anyone working out how allowing AI traffic changes their attack surface. It's the first of our releases on Iris. A Cyber Range Exercise is in the works, so watch this space. Start the lab: http://immersivelabs.online/labs/hunting-the-tools-iris-c2agent-log-analysisNew CTI Lab: Investigating a ClickFix Intrusion
ClickFix has become one of the most popular and successful social engineering techniques in the last decade. With multiple deployment options and a high success rate, cybercriminals are profiting considerably from unsuspecting victims in enterprise organizations and at the individual level. What is this about? In recent weeks, multiple intrusions have been reported in which cybercriminal actors compromised victims using ClickFix as an initial access method. ClickFix is a social engineering technique that convinces victims to copy malicious code to their clipboard and run it with administrator privileges under the guise of software updates, installations, computer repair, or troubleshooting. In this lab, we move away from Questions and Answers and towards action-based activities. This lab simulates realistic user-generated noise alongside realistic attacker actions, which are ingested by a SIEM. The user has to apply threat hunting techniques and then update the SIEM case management tool to reflect the discovered artefacts. Why is this critical for you and your team? The SOC is the beating heart of the security function. The SOC will often be the first individuals (second only to the victim) who will get any indication that a malicious attack has taken place - so it's important for the SOC to understand exactly how ClickFix intrusions work, what tactics attackers use, and how to threat hunt based on known techniques to quickly and accurately identify attacker behaviour. This lab will help users understand how ClickFix works and what goals an attacker might have after the ClickFix element of their intrusion concludes. Who is the content for? SOC Analyst Incident Responders Cyber Threat Intelligence Analysts Threat Researchers Threat Hunters Red Teamers Link to the lab: Investigating a ClickFix intrusionNew CTI Labs: Ruby Jumper Campaign: Analysis and Threat Actors: APT37
In February 2026, Zscaler ThreatLabz uncovered a sophisticated cyber espionage campaign orchestrated by APT37, assessed to be a North Korean state-sponsored threat actor. Dubbed "Ruby Jumper", this campaign represents a significant escalation in APT37's technical capabilities, specifically targeting air-gapped networks through weaponized removable media and abuse of legitimate cloud services. What is this about? Air-gapped networks are considered the final line of defense against threats to critical infrastructure, classified government systems, nuclear facilities, and high-value intellectual property. Ruby Jumper demonstrates that physical network isolation is no longer sufficient; sophisticated adversaries have developed reliable methods to bridge the air gap using removable media as bidirectional command channels. Why is this critical for you and your team? This campaign introduces novel techniques, including portable Ruby runtime environments for cross-platform malware execution, OAuth-based cloud C2 through legitimate services (Zoho WorkDrive, Google Drive, OneDrive), and self-propagating USB infection mechanisms. Understanding these TTPs is essential for organizations operating in air-gapped environments or defending against advanced persistent threats that leverage social engineering and trusted infrastructure to bypass traditional security controls. Who is the content for? Cyber Threat Intelligence Analysts Malware Reverse Engineers Threat Researchers Threat Hunters Link to the Labs: Threat Actors: APT37 APT37 - Ruby Jumper Campaign: AnalysisNew CTI Lab: CVE-2025-49113: Investigating a Roundcube RCE
In February 2026, the Cybersecurity and Infrastructure Security Agency (CISA) added CVE-2025-49113 to its Known Exploited Vulnerabilities (KEV) catalogue following exploitation in the wild. This critical vulnerability, which lay dormant in the Roundcube Webmail codebase for over a decade, allows authenticated attackers to achieve Remote Code Execution (RCE). With tens of thousands of instances exposed globally, particularly in government and higher education sectors, this flaw has become a primary target for both cybercriminal and state-sponsored groups. What is this about? CVE-2025-49113 is a high-impact PHP Object Deserialization vulnerability (CWE-502) with a CVSS score of 8.8. The flaw resides in how the application handles session data and URL parameters during file uploads. The attack vector focuses on: The Vulnerable Parameter: The _from parameter in program/actions/settings/upload.php lacks proper validation. The Logic Error: A bug in the Roundcube session parser allows an attacker to inject an exclamation mark (!) to corrupt session variables. The Gadget Chain: By manipulating the corrupted session, attackers can inject malicious PHP objects that leverage the Crypt_GPG library to execute arbitrary commands. Post-Auth Requirement: While exploitation requires authentication, attackers often pair this with credential harvesting or CSRF attacks to gain the initial foothold. Why is this critical for you and your team? This vulnerability represents a long-standing vulnerability that many organizations have failed to patch to date: Execution: Exploitation is notoriously difficult to detect via traditional Web Application Firewalls (WAFs) due to the nature of PHP object injection within session handling. Admins are largely relegated to access logs and HTTP logs to identify evidence of intrusion Federal Mandate: CISA has directed all Federal Civilian Executive Branch (FCEB) agencies to remediate this vulnerability by March 13, 2026, underscoring the immediate risk to national security infrastructure. Who is the content for? Security Analysts System Administrators Threat Researchers Threat Hunters Link to the lab: CVE-2025-49113: Investigating a Roundcube RCENew CTI Lab: 7-Zip Installer (Proxy Node Campaign): Analysis
In February 2026, security researchers across different organizations exposed a long-running malware distribution campaign targeting users of the popular 7-Zip archiving utility. Operating for an extended period, threat actors behind the operation registered the convincing lookalike domain 7zip[.]com , closely mimicking the legitimate 7-zip.org to distribute trojanized installers that silently convert victims' machines into residential proxy nodes. What is this about? Brand impersonation attacks represent a critical threat vector where attackers exploit user trust rather than software vulnerabilities. In this campaign, operators created a sophisticated intrusion using a fake 7zip[.]com domain that mirrors the structure and content of the official site, the malicious installer carries a now-revoked code-signing certificate from "JOZEAL NETWORK TECHNOLOGY CO., LIMITED," and victims receive a fully functional copy of 7-Zip that deploys malicious payloads onto the victim machine. These malicious Golang binaries establish persistence, manipulate firewall rules, and transform victim machines into nodes for a residential botnet. Why is this critical for you and your team? As security teams increasingly focus on advanced persistent threats and zero-day exploitation, this campaign demonstrates how attackers achieve persistent access through social engineering and trust exploitation. Users downloading software from what appears to be a legitimate source, particularly when following online tutorials or search engine results, and bypassing traditional security awareness training. The malware's use of code-signed binaries, legitimate system directories, and SYSTEM-level service persistence means it evades many endpoint security controls designed to catch obvious malware. Understanding this infection chain and learning to threat hunt for these artefacts is essential for detecting similar tactics in your environment. Who is the content for? Security Analysts Threat Researchers Threat Hunters Here is a link to the lab: 7-Zip Installer (Proxy Node Campaign): AnalysisNew CTI Lab: Lotus Blossom Notepad ++ Campaign: Analysis
In January 2026, threat researchers at Rapid7 detailed a sophisticated supply chain attack targeting the Notepad++ update mechanism. Between July and October 2025, attackers compromised the project’s distribution infrastructure to deliver a custom, undocumented backdoor dubbed Chrysalis. By intercepting update requests, the threat actor distributed malicious NSIS installers to a targeted set of victims across Southeast Asia and Australia. What is this about? Supply chain compromises represent one of the most dangerous threat vectors today. In this campaign, the Chinese state-sponsored group Lotus Blossom (also known as Billbug or Thrip) hijacked a trusted software update pipeline. The attack involves complex DLL sideloading techniques—abusing a renamed Bitdefender binary to execute a multi-layered encrypted payload. Once the Chrysalis backdoor is active, it provides the attackers with persistent, feature-rich remote access to the victim's environment. Why is this critical for you and your team? As organizations rely on legitimate third-party utilities like Notepad++, trust in the update process is paramount. This intrusion highlights how state-sponsored actors can weaponize that trust to bypass perimeter defences. Understanding the Chrysalis infection chain—from the initial NSIS installer to the triple-layer decryption of its C2 configuration—is vital for detecting similar "living-off-the-land" and sideloading tactics in your own network. If your team manages software deployments or monitors developer environments, you must be cognisant of how attackers leverage legitimate, signed binaries to mask malicious behaviour. This lab provides a deep dive into the specific obfuscation and persistence strategies used by one of the region's most persistent threat groups. Who is the content for? Security Analysts Threat Researchers Here is a link to the lab: Lotus Blossom Campaign: AnalysisNew CTI Lab: CVE-2026-23744 (MCPJam RCE): Offensive
On January 16, 2026, advisories were released covering a critical vulnerability in MCPJam Inspector, the local-first development platform for MCP servers. The Latest version, 1.4.2 and earlier, is vulnerable to a remote code execution (RCE) vulnerability, a trivial yet highly impactful vulnerability that allows an attacker to send a crafted HTTP request that triggers the installation of an MCP server, leading to RCE. What is this about? Model context protocol (MCP) has become more popular as a way to connect applications and services together that use AI, such as connecting tools to your OpenAI account, so the AI can help you work with the tool, perform tasks on your behalf, or work as webhooks between tools. MCPJam is an example of a tool that makes these processes easier and more convenient. Why is this critical for you and your team? As AI adoption across industries and sectors rises, products and services have been released to help people interact with AI pipelines. With MCPJam and tools like it, you can test and develop MCP (model context protocol) servers, emulate deployments, and debug your workflow, making your entire MCP development pipeline much smoother.If you're using any tools like this, where you share you API keys and other sensitive data with the tool, you need to be cognisant of the risks that these tools carry, as many others are vulnerable to basic misconfigurations that can lead to serious impacts. Who is the content for? Penetration Testers Security Analysts Incident Responders Here is a link to the lab: CVE-2026-23744 - MCPJam: Offensive This application has no logging available at all, so no Defensive Variant of this labNew CTI Lab: CVE-2026-21858 (n8n RCE): Offensive
On January 7, 2026, Cyera Research Labs released an advisory for "Ni8mare," a critical unauthenticated remote code execution vulnerability (CVE-2026-21858) in n8n with a CVSS score of 10.0. The flaw stems from a "Content-Type Confusion" bug in the Form Webhook node, which allows attackers to override internal file paths, thereby enabling the arbitrary disclosure of sensitive data, including database.sqlite and the system's unique encryption key. This vulnerability can be exploited to forge administrative sessions and achieve full system takeover. What is this about? n8n has had a lot of vulnerabilities over the last year or so, in particular, these vulnerabilities have required authentication to exploit, meaning the attacker would already need access to the n8n server to leverage the vulnerability. This critical vulnerability has a CVSS score of 10.0, and attackers can achieve unauthenticated remote code execution - making it a poignant discussion point for potential future vulnerabilities and attacks, given that n8n will likely receive more attention from vulnerability researchers and threat actors alike. Why is this critical for you and your team? n8n is very popular with organization and the wider community alike, with over 70,000 active instances exposed to the internet; there is a reasonably wide attack surface to be exploited. If you or your team uses n8n, and there is a reasonably high probability that you do (for example in human resources, project planning, news feeds etc) then learning about and mitigating this vulnerability is essential to protect yourself against attacks. Who is the content for? Penetration Testers Security Analysts Incident Responders Here is a link to the lab: CVE-2026-21858 (n8n RCE): OffensiveNew CTI Lab: CVE-2025-55182 (React - Next.js)
On December 3, 2025, the cybersecurity world received news of a critical vulnerability in the React 19 ecosystem. This critical flaw, tracked as CVE-2025-55182 with a CVSS score of 10.0, affects React Server Components (RSC). A major issue, this flaw allows unauthenticated attackers to achieve Remote Code Execution (RCE) on vulnerable servers by sending a specially crafted HTTP request. AI Hallucination Within the first 24 hours of the vulnerability being announced, a POC was published to GitHub, which looked convincing and, when tested, appeared to achieve the goal successfully, resulting in Code Execution. It turned out that this POC, which was picked up and circulated by researchers and social media, was actually an AI Hallucination. The AI had crafted a deliberately misconfigured and vulnerable server and a POC that appeared to match the requirements of the exploit, but only actually triggered the misconfiguration. What is this about? CVE-2025-55182 is a critical Insecure Deserialization vulnerability. It affects React Server Components (RSC) within the React 19 ecosystem. The flaw is located in the server-side logic that handles the React Flight protocol, which is used for client-to-server interactions, specifically Server Functions or Server Actions. An unauthenticated attacker can execute a specially crafted HTTP request containing a malicious, serialized payload. The vulnerable server-side code fails to validate this payload, allowing the attacker to achieve remote code execution on the server. Why is this critical for you and your team? This critical vulnerability has a CVSS score of 10, is fairly trivial to exploit, and has significant impacts when successfully exploited, given that its impact includes unauthenticated remote code execution. If your team uses React, React Server Components (RSC), or similar, are at risk. This flaw impacts the standard, default configurations of high-profile frameworks like the Next.js App Router, which many organizations rely on for building high-performance sites. Who is the content for? Security Analysts Penetration Testers Incident Responders Vulnerability Management Teams Here is a link to the lab: CVE-2025-55182 (React - Next.js)New CTI Lab: Shai-Hulud 2.0: Analysis
In late November/early December 2025, a set of critical software supply chain intrusions took place when the highly dangerous Shai-Hulud 2.0 worm was used to steal GitHub, Cloud, and other credentials and secrets by gaining access to developer machines through the use of a malicious npm package installation. What is this about? By abusing the inherent trust in the npm ecosystem, Shai-Hulud guarantees execution during the crucial preinstall phase, effectively bypassing many traditional security scans that only review code after installation. Once running, the payload launches a concurrent, parallel attack across your environment: it hunts for local credentials, attempts to steal highly privileged temporary cloud tokens via the Instance Metadata Service (IMDS), and, most critically, can automatically inject itself into every other package the victim maintains on their machine. Why is this critical for you and your team? npm is massively popular, and many of the affected packages are widely used in software development and deployment. Shai-Hulud 2.0 is a devastating self-replicating worm that weaponizes your supply chain to steal highly privileged cloud credentials (IMDS) and establish a permanent C2 backdoor via GitHub Actions if the threat actor decides to set that up. Given the importance of npm packages to developers, customers from any organisation, and across all sectors, it is essential that they understand how this intrusion works to prevent their credentials and secrets from being stolen. Who is the content for? Security Analysts Incident Responders Software Developers/Secure Development teams Cloud Engineers Vulnerability Management Teams Here is a link to the lab: Shai-Hulud 2.0: Analysis
