China-linked hackers use DeepSeek AI to hit 200 organizations in 14 countries

A coordinated global campaign

A hacking group believed to be linked to China recently conducted a massive cyberattack targeting more than 200 organizations across 14 different countries. According to a report from CrowdStrike, the campaign took place over an 11-day period starting on the 17th of last month. The scope of the operation was notably broad, reaching across multiple continents and industries.

The targets included a diverse range of high-value entities, such as American firms in the defense and semiconductor sectors, as well as companies involved in pharmaceuticals, nuclear energy, and finance. The hackers also aimed at government bodies, specifically hitting the foreign ministries and presidential offices of various European countries. This wide net suggests a strategic effort to gather intelligence from both private and public sectors simultaneously.

CrowdStrike confirmed that the group successfully gained access to at least nine of these organizations. Once inside, the attackers planted a malicious backdoor known as BLOODALCHEMY to facilitate further intrusion. The report also noted that telecom carriers in the United States, Belgium, and the Philippines were compromised during the campaign.

AI agents as the primary hackers

The most significant aspect of this incident is that autonomous AI agents, not human operators, performed the majority of the hacking work. The hackers utilized DeepSeek, a Chinese large language model, and ARTEX, which was originally an open-source penetration testing tool. By combining these technologies, the attackers were able to divide roles among multiple AI agents to run tasks in parallel.

These agents handled various stages of the attack, including scanning for vulnerabilities, system intrusion, internal reconnaissance, and privilege theft. The hackers issued commands to the AI directly in Chinese and only intervened at key decision points to steer the direction of the operation. This setup allowed a small team of humans to manage a much larger volume of attacks than would be possible manually.

The technical scale of the AI's involvement was substantial. In one instance, the AI agent responsible for formulating an attack plan processed roughly 1.4 billion input tokens to complete its task. This level of automation allows attackers to move through complex systems with a speed and consistency that is difficult for traditional security measures to keep up with.

Specific targets and tactics

The campaign showed a distinct regional pattern of damage across Latin America, Europe, Asia, and the Middle East. In one European country, the attackers hit a concentrated list of targets, including the foreign ministry, presidential office, and prime minister's office. They also targeted the justice, finance, defense, and intelligence agencies within that same nation.

In the United States, the hackers focused on 19 institutions within the pharmaceutical, semiconductor, defense, nuclear, and financial sectors. The tactics used varied depending on the target. For example, at one US pharmaceutical company, an AI agent linked several different flaws together to reach a cloud system and then took over administrator privileges.

Other specific tactics included the installation of web shells to obtain root-level administrator access in container environments. At a government agency in Greece, the attackers used previously leaked credentials to forge administrator authentication. This allowed them to gain unauthorized access to internal networks and email systems, demonstrating how AI can be used to exploit existing data more efficiently.

The evolution of cyber threats

Security industry experts view this incident as a major evolution in how AI-based cyberattacks are conducted. It confirms that AI can move beyond penetrating single systems to targeting multiple countries and institutions at the same time. This shift means that even a hacking group with only a few members can strike far more targets than they could previously.

The use of ARTEX is particularly notable because it was designed as a tool for finding security vulnerabilities, not as a weapon. However, attackers have successfully repurposed this open-source tool to conduct actual hacking. It has been seen in previous attacks targeting the financial sector in South Korea, showing a pattern of weaponizing security tools.

CrowdStrike projects that China-linked hacking groups are highly likely to continue combining AI agent frameworks with commercial models over the next 12 months. As AI lowers the barrier to entry for creating sophisticated attacks, the speed of the defenders' response becomes a critical variable. Organizations will need to improve their detection capabilities to counter these automated, high-speed intrusions.

What this means for security

For the average organization, this news highlights a growing reality where cyber threats are becoming more automated and harder to detect. The ability of AI to perform complex multi-stage attacks at high speed means that traditional, human-led defense strategies may struggle to keep pace.

The fact that attackers can now use AI to scale their operations suggests that the volume of threats will likely increase. Companies must prioritize identifying automated behaviors rather than just looking for known signatures of human-led attacks.

Ultimately, this incident underscores the need for a more proactive security posture. As AI tools become more accessible to malicious actors, the focus must shift toward rapid detection and automated response to mitigate the damage of these high-speed, AI-driven campaigns.