State-Aligned Hackers Automate Intrusions at Industrial Scale Using Open-Source AI

The Attack Vector

China state-aligned threat actors breached ConnectWise's cloud backup infrastructure and exfiltrated application credentials used to establish trust relationships between the IT management platform and its corporate customers. The stolen tokens allowed attackers to impersonate legitimate users within Microsoft-hosted environments, granting access to email, file storage, and other sensitive resources across approximately 2,250 organisations.

ConnectWise, which provides remote access and IT management tools to managed service providers and enterprises, suspended access to its cloud backup system following discovery of the intrusion. As of late October 2026, the company stated it was working to restore service whilst investigating the scope of the compromise.

The breach underscores the systemic risk embedded in software supply chains: a single point of compromise at a service provider can cascade across thousands of downstream customers, many of whom may lack visibility into the initial vector.

AI-Accelerated Intrusion

What distinguishes this incident is not the initial breach but the velocity of subsequent exploitation. Researchers at Microsoft observed attackers deploying AI agents to automate credential validation and environment enumeration at nearly 3,000 sign-in attempts per minute. At that cadence, defenders face what Microsoft characterised as minimal time to detect anomalies, correlate signals, and mount a response before data exfiltration or lateral movement occurs.

The attackers employed ARTEX, an open-source AI framework designed to work with large language models for penetration testing and adversarial simulation. Originally built to stress-test cyber defences, ARTEX allows operators to script complex workflows that adapt to environment feedback, automate reconnaissance, and optimise exploitation paths.

ARTEX is not itself malicious; it is a tool used by red teams and security researchers to simulate sophisticated attacks. Its availability as open-source software, however, means state-aligned groups can integrate it into operational playbooks without needing to develop proprietary automation from scratch. The framework's flexibility and speed make it particularly effective in scenarios where attackers already possess valid credentials and need to pivot rapidly across multiple targets.

Regional Precedent

ARTEX has surfaced in at least one other high-profile campaign. In recent months, threat actors used the same framework in a compromise targeting South Korea's financial sector. That incident exploited institutions with inadequate security postures, including a widespread absence of multi-factor authentication. The South Korean intrusions demonstrated ARTEX's capacity to scale attacks horizontally: once an initial foothold is secured, the AI agent can enumerate and exploit additional weaknesses across interconnected systems faster than human operators.

The recurrence of ARTEX in both the ConnectWise incident and the South Korean campaign suggests that state-aligned groups are systematically incorporating AI-driven automation into their operational toolkit. This shift has implications for defenders, who must now contend with attack tempos that outpace traditional detection and response cycles.

The Industrialisation of Intrusion

The ConnectWise breach exemplifies a broader trend: the industrialisation of cyber intrusion. Where earlier campaigns relied on manual reconnaissance and bespoke exploit chains, attackers now leverage automation to achieve scale and speed previously unattainable. AI frameworks like ARTEX lower the operational overhead of mass exploitation, enabling a single team to target thousands of organisations in parallel.

This evolution is not confined to credential abuse. AI-assisted reconnaissance can identify vulnerable endpoints, enumerate network topologies, and prioritise high-value targets with minimal human intervention. In environments where defenders rely on anomaly detection, the sheer volume and velocity of automated activity can overwhelm alert queues, rendering traditional security operations centres less effective.

The use of open-source AI tools also complicates attribution and mitigation. Because ARTEX and similar frameworks are publicly available, they leave fewer forensic artefacts that tie an intrusion to a specific threat actor. Defensive measures must therefore focus less on identifying the tool and more on detecting the behavioural patterns that accompany automated exploitation, such as rapid credential enumeration or anomalous sign-in velocity.

Implications for Managed Service Providers

The ConnectWise incident places renewed scrutiny on managed service providers and IT management platforms. These entities occupy a privileged position within customer networks, often holding credentials and access rights that span entire environments. When such a platform is compromised, the blast radius extends well beyond the provider itself.

For enterprises that rely on MSPs, the breach raises uncomfortable questions about third-party risk management. Many organisations lack visibility into how their MSPs store credentials, segment customer environments, or monitor for anomalous activity within shared infrastructure. The ConnectWise case suggests that even well-established providers can fall victim to sophisticated intrusions, and that customers must assume breach scenarios when designing their own defences.

Multi-factor authentication remains a foundational control. Had affected organisations enforced MFA on all external access, the stolen credentials would have been significantly less useful to attackers. The fact that nearly 3,000 sign-in attempts per minute succeeded implies that many environments lacked this basic safeguard.

Defensive Adaptation

Defenders face a challenge: how to detect and respond to AI-accelerated attacks when the attack tempo exceeds human decision-making speed. Traditional security operations rely on analysts triaging alerts, investigating anomalies, and coordinating response actions. When attackers can enumerate thousands of targets in minutes, that workflow breaks down.

One response is to deploy automated defences that match the speed of automated attacks. Behavioural analytics engines that flag anomalous sign-in velocity, geolocation mismatches, or unusual resource access patterns can trigger automatic lockouts or step-up authentication challenges without waiting for human review. Organisations that have invested in security orchestration, automation, and response platforms are better positioned to counter AI-driven intrusions.

Another is to assume that credentials will be stolen and design environments accordingly. Zero-trust architectures that enforce least-privilege access, segment workloads, and continuously verify identity can limit the damage from compromised credentials. If an attacker gains access to an environment but finds that lateral movement is blocked by micro-segmentation and that data access requires continuous re-authentication, the value of the initial foothold diminishes.

The Open-Source Dilemma

ARTEX's role in this incident highlights a tension within the security community. Open-source offensive tools are invaluable for red teams, researchers, and organisations seeking to test their own defences. They democratise access to sophisticated techniques and enable smaller teams to simulate advanced adversaries.

Yet the same openness that benefits defenders also equips attackers. When a framework designed to stress-test cyber defences is adopted by state-aligned groups, the line between legitimate research and operational exploitation blurs. The security community has long grappled with this dual-use dilemma, and no consensus has emerged on whether or how to restrict access to such tools.

In practice, restricting access to open-source frameworks is difficult to enforce and may simply drive development underground. A more pragmatic approach is to assume that adversaries will use the best available tools and focus defensive investment on detection and response capabilities that can counter those tools regardless of their provenance.

Looking Forward

The ConnectWise breach will not be the last incident in which AI automation plays a central role. As AI frameworks become more sophisticated and easier to deploy, the barrier to entry for large-scale, high-velocity intrusions will continue to fall. State-aligned groups, criminal syndicates, and opportunistic actors will all benefit from this trend.

For defenders, the imperative is clear: automate defences, enforce strong authentication, segment environments, and assume that any credential can be compromised. The era of manual, low-tempo intrusions is giving way to one in which attackers operate at machine speed. Organisations that fail to adapt will find themselves unable to keep pace.