Key Takeaways
- Taiwan’s Ministry of Digital Affairs detected an overseas AI‑assisted cyber‑attack on government systems that began on 20 July, describing it as a “first‑of‑a‑kind breach.”
- The intrusion was identified by the Israeli AI firm Dream, which reported that attackers used open‑source AI agents to create an autonomous hacking tool that compromised at least 85 government accounts and exfiltrated over 2,500 personnel records.
- The attack subsequently spread to Taiwan’s nuclear safety agency and at least seven energy companies, highlighting the potential impact on critical infrastructure.
- While Taiwanese officials did not name a perpetrator, the use of Simplified Chinese in the hackers’ internal communications led analysts to assess a high probability of China‑linked involvement, consistent with Beijing’s broader “hybrid warfare” tactics.
- Chinese cyber‑activity against Taiwan’s key sectors rose 6 % in 2025 to an average of 2.63 million daily attacks, some of which were synchronized with military drills.
- Taiwan’s investigation confirmed the attack’s overseas origin, its hybrid manual‑AI methodology, and has responded by issuing protective guidelines and strengthening real‑time system monitoring to block similar threats early.
Detection of AI‑assisted cyber‑attacks
Taiwan’s Ministry of Digital Affairs (MDA) announced that its cybersecurity monitoring units uncovered an “abnormal attack” targeting multiple government agencies that commenced on 20 July 2025. The MDA’s National Institute of Cyber Security issued a series of warning alerts while it investigated the incident, which officials characterized as a novel, first‑of‑a‑kind breach because it leveraged artificial intelligence to automate and scale hostile operations. The discovery came amid heightened vigilance over Taiwan’s digital defenses, reflecting growing concerns that adversaries are increasingly integrating AI tools into traditional cyber‑espionage playbooks.
Role of the Israeli AI firm Dream
The Financial Times quoted Dream, an Israeli AI security company that first detected the intrusion, as saying the attackers employed open‑source AI agents to construct an autonomous hacking tool behaving like a coordinated cyber team. Dream described the tool as a “first‑of‑a‑kind breach” because it combined machine‑learning‑driven reconnaissance, vulnerability identification, and exploitation without constant human direction for each step. According to Dream, the compromise extended beyond initial infiltration, affecting a broad swath of Taiwan’s governmental and critical‑infrastructure networks.
Scope of the compromised data
Dream reported that the AI‑assisted tool compromised at least 85 government user accounts, from which it extracted more than 2,500 personnel records before the attack pivoted to Taiwan’s nuclear safety agency and at least seven energy companies. The exfiltration of sensitive personnel data raised alarms about potential insider threats, while the subsequent targeting of nuclear safety and energy sectors underscored the attackers’ intent to disrupt essential services and gather strategic intelligence about Taiwan’s energy security and nuclear capabilities.
Taiwan’s broader security concerns
In recent years Taiwan has repeatedly warned of what it describes as China’s “hybrid warfare”—a blend of daily military drills near the island, disinformation campaigns, and cyber‑attacks aimed at pressurizing the democratically governed island to accept Beijing’s sovereignty claims. Although Taiwanese officials refrained from directly attributing the July incident to China, the pattern aligns with Beijing’s broader strategy of using cyber tools to complement conventional pressure tactics, seeking to erode Taiwan’s resilience without triggering outright military confrontation.
Suspicions of China‑linked involvement
The Financial Times noted that, while Dream did not assign the attack to a specific threat actor, the presence of Simplified Chinese in the hackers’ internal communications led analysts to judge a high probability that the operators were connected to China. This linguistic clue, combined with the timing and nature of the intrusion, reinforced existing assessments that Chinese state‑backed or affiliated groups are increasingly experimenting with AI‑enhanced techniques to achieve strategic objectives against Taiwan.
Scale of Chinese cyber activity
Supporting the suspicion of Chinese involvement, Taiwan’s National Security Bureau reported in January 2025 that cyber‑attacks on Taiwan’s key infrastructure—including hospitals, banks, and energy providers—had risen 6 % year‑on‑year to an average of 2.63 million attacks per day. The bureau added that a notable portion of these hacks were synchronized with Chinese military drills, exemplifying the “hybrid threat” model where cyber operations are timed to amplify the psychological and operational impact of conventional shows of force.
Investigation findings and response measures
The MDA’s investigation concluded that the attack exhibited clear characteristics of an overseas source, employing a hybrid approach that blended manual operator decisions with AI agent‑assisted actions, exemplified by tools such as “Open Claw.” The ministry affirmed that the relevant attack sources, methods, and scope of impact had been fully investigated and that affected units had completed their remediation steps. In response to this emerging threat, Taiwan has issued protective guidelines for government and critical‑infrastructure entities and has intensified real‑time system monitoring to detect and block similar AI‑driven intrusions at an early stage.
Attribution ambiguity and official silence
Despite the compelling indicators, the MDA’s statement deliberately avoided naming China as the perpetrator, and China’s Taiwan Affairs Office did not immediately respond to a request for comment on the attack. Similarly, Taiwanese authorities and Chinese officials declined to comment on the earlier Financial Times report that had highlighted the suspected China link. This diplomatic restraint reflects the sensitivities surrounding public attribution in cyber‑conflict, where outright accusations can escalate tensions while concrete proof remains challenging to produce in the public domain.
Expert perspective on AI‑assisted hacking
Cris Thomas, a researcher and security advocate at the code‑security firm Semgrep, observed that the incident illustrates how rapidly AI‑assisted hackers can achieve their objectives once a human operator defines the target and goals. Thomas cautioned against overstating the autonomy of AI agents, emphasizing that “there’s still a human in there somewhere… somebody had to choose who to attack, had to establish an objective and give it a directive.” He noted that while AI can accelerate reconnaissance and exploitation, the strategic intent and ultimate decision‑making remain firmly human‑driven, underscoring the importance of focusing defenses on both the technological and human elements of cyber threats.
Implications for future cyber defense
The Taiwan incident signals a watershed moment in the evolution of cyber warfare, demonstrating that adversaries can now harness widely available open‑source AI models to automate stages of an attack that previously required extensive manual labor. As top AI labs continue to release powerful models capable of rapid vulnerability identification and exploitation, nations must adapt by investing in AI‑aware threat detection, sharing indicators of compromise across sectors, and cultivating hybrid defense teams that blend machine‑learning analytics with seasoned human analysts. Taiwan’s proactive issuance of protective guidelines and enhanced monitoring exemplifies a forward‑looking approach that other jurisdictions may emulate to safeguard critical infrastructure against the next generation of AI‑enabled cyber threats.