Cybersecurity researchers have uncovered a large international cyber espionage campaign in which a China-linked hacking group used artificial intelligence (AI) agents to target more than 200 organisations across at least 14 countries. The operation, attributed to a group tracked as VAPOR PANDA, involved an AI-powered penetration testing framework called ARTEX, combined with large language models to automate different stages of cyberattacks.

The campaign began on October 17, 2026, and targeted government institutions, telecommunications providers, pharmaceutical companies, semiconductor organisations, defence entities, and other high-value targets. The findings, published by cybersecurity company CrowdStrike on October 29, highlight how attackers can use AI agents to coordinate multiple operations simultaneously rather than carrying out every task manually.

What Happened in the AI-Powered Cyberattack?

CrowdStrike’s threat intelligence team identified a campaign in which VAPOR PANDA used a single ARTEX controller to coordinate attacks against organisations in at least 14 countries over an 11-day period.

The targets included foreign ministries in Latin America, Europe, Asia, and the Middle East. The campaign also involved organisations in the United States, including businesses and institutions connected to pharmaceuticals, semiconductors, defence, nuclear research, and financial services.

Telecommunications providers in the United States, Belgium, and the Philippines were also among the targets.

CrowdStrike reported post-exploitation activity at nine organisations. In these cases, the attackers gained access, deployed malicious software, and continued activities such as credential discovery, network scanning, and reconnaissance.

It is important to distinguish between the more than 200 organisations targeted and the nine organisations where CrowdStrike confirmed post-exploitation activity. The figures do not mean that every targeted organisation was successfully compromised.

What Is ARTEX and How Does It Work?

ARTEX is an AI-powered penetration testing framework designed to automate security testing activities. It uses multiple AI agents that can work together to perform tasks such as reconnaissance, vulnerability identification, and further actions within a target environment.

In legitimate security testing, tools like these can help authorised professionals identify weaknesses before criminals exploit them.

However, the same automation capabilities can be misused by threat actors.

ARTEX uses a main agent that coordinates smaller planning and worker agents. These agents share information, track findings, and work towards objectives assigned by the operator.

For the VAPOR PANDA campaign, CrowdStrike reported that ARTEX was used alongside the Chinese-developed DeepSeek large language model and another commercial AI model whose identity was not confirmed.

This combination allowed the attackers to pursue multiple intrusion paths at the same time. Rather than waiting for a human operator to complete every individual task, the agents could carry out several activities in parallel and adapt when one approach failed.

How Did Hackers Use AI Agents During the Attacks?

The campaign demonstrates how AI can help attackers automate several stages of an intrusion.

1. Automated reconnaissance

The agents examined target environments to identify systems, services, and potential weaknesses. This helped the attackers gather information about organisations before deciding which access paths to pursue.

2. Finding and exploiting security weaknesses

CrowdStrike described cases in which the agents used existing weaknesses in web applications, cloud services, and other systems to expand access.

The framework could also switch to alternative approaches when an attempted route did not work.

3. Credential discovery

The attackers used the agents to search for authentication information, including credentials and keys that could provide access to additional systems.

In one reported case, historical authentication keys found in an organisation’s directory export were used to access mailboxes. This demonstrates why old credentials and unrotated keys can remain security risks even when they are no longer actively used by employees.

4. Moving through connected systems

After gaining access, the agents investigated connected systems and attempted to move into other parts of the network.

The reported activity involved cloud infrastructure, container environments, identity systems, email services, and VPN connections.

5. Deploying malicious software

CrowdStrike reported that VAPOR PANDA deployed a backdoor called BLOODALCHEMY in some compromised environments.

A backdoor is malicious software that can provide an attacker with continued access to a system. In this campaign, the malware was disguised using files and names associated with legitimate software, making investigation more difficult.

These activities show that AI did not replace the attacker’s tools or objectives. Instead, it helped coordinate and accelerate techniques that cybercriminals and espionage groups have used for years.

Which Organisations Were Targeted?

The campaign focused on organisations with access to strategically important information, infrastructure, and technology.

The range of targets suggests that the operation was focused on gathering access and intelligence across several strategically important sectors.

However, targeting an organisation does not automatically mean that its data was stolen or that all its systems were compromised.

Why Is This Attack Different From Traditional Cyberattacks?

Traditional cyberattacks often depend heavily on human operators to perform reconnaissance, review information, select targets, and decide what to do next.

AI agents can reduce the amount of manual work involved in these activities.

In this campaign, ARTEX coordinated multiple agents that could pursue different attack paths simultaneously. CrowdStrike reported that a single intrusion consumed approximately 1.4 billion input tokens across the planning agents, illustrating the scale of automated activity involved.

The main difference is not necessarily that AI creates entirely new hacking techniques. It can make existing techniques faster, easier to coordinate, and more scalable.

A relatively small team may therefore be able to investigate more targets and perform more operations than would be practical through manual work alone.

What Should Organisations Do to Protect Themselves?

The campaign highlights the importance of securing identities, cloud environments, endpoints, and internet-facing systems.

Organisations should consider the following measures.

1. Strengthen identity and access controls

Use multi-factor authentication, restrict administrative privileges, and review accounts with access to sensitive systems. Remove unnecessary permissions and investigate unexpected account changes.

2. Rotate exposed credentials and keys

Old authentication keys, service credentials, and signing keys should not remain active indefinitely. If historical backups or directory exports may have exposed them, organisations should assess whether rotation is necessary.

3. Secure cloud infrastructure

Review cloud permissions, application identities, and service accounts. Restrict administrative command execution and ensure that cloud resources only receive the permissions required for their functions.

4. Patch internet-facing systems

Keep servers, web applications, network appliances, and other externally accessible systems updated. Vulnerabilities in exposed services can provide attackers with an initial route into an organisation.

5. Monitor unusual activity

Security teams should monitor unexpected authentication attempts, unusual network scanning, suspicious application activity, and attempts to access systems outside normal working patterns.

6. Protect endpoints and investigate gaps

Ensure that endpoint detection and response tools cover servers and workstations. Investigate devices that lack security monitoring or show signs of unexpected software execution.

What Does This Mean for the Future of Cybersecurity?

The VAPOR PANDA campaign is an important example of how artificial intelligence is changing the cybersecurity landscape.

AI agents can help defenders analyse threats, automate repetitive security checks, and identify weaknesses. At the same time, attackers can misuse similar capabilities to increase the speed and scale of intrusions.

This creates a challenge for security teams. Defenders must be able to identify suspicious activity quickly, investigate multiple systems, and respond before attackers establish deeper access.

The campaign also shows why AI security cannot be separated from basic cybersecurity practices. Strong identity management, protected credentials, regular patching, network monitoring, and tested incident response procedures remain essential.

Final Thoughts

The AI-powered cyber espionage campaign attributed to VAPOR PANDA demonstrates how automated agents can support large-scale attacks against organisations across different countries and industries. By combining ARTEX with large language models, the group was able to coordinate multiple intrusion activities and target more than 200 organisations in an 11-day period.

CrowdStrike confirmed post-exploitation activity at nine organisations, including the deployment of the BLOODALCHEMY backdoor and further attempts to discover credentials and explore internal networks.

The key lesson is that AI is making cyber operations more scalable, but it does not remove the need for human oversight or replace the importance of strong security controls. Organisations must improve their ability to detect unusual activity, protect privileged access, and respond quickly to threats that can operate at machine speed.