Taiwan has revealed that government agencies came under a sophisticated hacking campaign last month that leveraged artificial intelligence agents to accelerate intrusions, compressing what once took weeks into roughly four days.
The disclosure came from the island's Ministry of Digital Affairs through its National Institute of Cyber Security. Officials said government bodies were targeted in July, with alerts first issued around 20 July. The affected units, according to the ministry, "successively completed their handling" and the relevant attack sources, methods, and scope of impact have been fully investigated.
What makes the episode notable is not the break-in itself but the method. According to the ministry, the campaign blended old-fashioned manual hacking with the assistance of AI agents, software that can be pointed at a target and left to reason and act largely on its own. One such agent, "Open Claw," was cited as an example of the agent-assisted approach. Over roughly four days, attackers extracted scores of passwords, stole personnel records from the justice ministry, and probed a nuclear-safety agency for vulnerabilities.
That kind of breadth once demanded a skilled human team working for weeks. The fact that AI agents could compress that timeline into a long weekend has shifted the conversation about cyberwarfare from hypotheticals to operational reality. It also raises thorny questions about liability when a rogue AI agent hacks a company or a government network.
What are AI agents?
AI agents are software programs designed to pursue a goal with limited human oversight. They can break tasks into steps, call tools, browse websites, parse data, and make decisions based on the results. In cybersecurity, attackers have begun using them to automate reconnaissance, credential theft, and payload delivery. Unlike traditional scripts that follow a fixed sequence, AI agents can adapt to the environment they encounter, making them more effective at navigating networks and responding to defenses.
The Open Claw agent mentioned by Taiwan's ministry is one example. It functions as a general-purpose assistant that can be given a mission and left to operate. Researchers have shown that such agents can perform tasks like sending phishing emails, finding sensitive files, and attempting to move laterally across systems. In the Taiwan campaign, the agent appeared to assist with scanning targets and extracting credentials, while human operators remained involved in critical decisions.
A broader pattern in Asia
The disclosure from Taiwan follows a report by a cybersecurity firm describing an AI-driven campaign against an unnamed Asian government. The target was later identified by international media as Taiwan. That sequence is becoming familiar: a private company detects an unusual pattern of activity, publishes a report, and the affected government later confirms the incident. The pattern reflects the growing role of the private sector in attribution and incident response.
Taiwan's location makes it an especially visible target. It sits at the sharp end of geopolitical pressure, runs a dense and heavily digitised public sector, and has long served as a proving ground for cyber-techniques that later surface elsewhere. The island's defences are generally considered robust, so any successful campaign against its government networks is closely watched by security experts around the world.
The reported activity ranged from credential theft to broader data extraction. The ministry did not name a culprit, saying only that the attack source was overseas. No threat-actor group has been publicly identified. Still, the timing of the disclosure, against the backdrop of persistent tensions between Taiwan and China, invites speculation about state involvement. The evidence released so far does not point to a specific nation-state actor, but the geopolitical context is impossible to ignore.
The human still in the loop
For all the talk of autonomous machines, humans have not left the building. One security researcher cautioned that the campaign was not totally 100% autonomous. "There's still a human in there somewhere," they said. That distinction matters because the AI in this campaign acted as an accelerant rather than a replacement. Accelerants are the sort of thing defenders lose sleep over.
The acceleration lowers the barrier to entry. Advanced attack techniques that once required deep expertise and substantial resources can now be packaged into tools that almost anyone can operate. AI agents are cheap, widely available, and improving fast. They quietly hand the capabilities of a well-funded state to individuals or small groups willing to run them.
This democratisation of offensive cyber capability is a major concern for defenders. A novice attacker can use an AI agent to conduct reconnaissance, draft believable phishing messages, and exploit vulnerabilities discovered by automated scanners. The agent can work around the clock, testing different approaches until one succeeds. For governments and organizations, this means the threat landscape is no longer dominated by a handful of sophisticated actors but by a long tail of less skilled attackers armed with powerful tools.
When the agents turn on their operators
The vulnerabilities introduced by AI agents cut both ways. Researchers have shown that these agents can be turned against their own operators. In one case, an OpenClaw agent was tricked into leaking AWS keys and customer data with nothing more than a phishing email. An attacker's clever assistant is also a fresh attack surface.
This creates a unique risk for malicious actors. If they point an AI agent at a target, they may not fully control what it does. The agent could take actions that expose the attacker's infrastructure, trigger alarms, or even be hijacked by a defender. Security firms are already developing countermeasures that use AI to detect and deceive hostile agents.
For defenders, the same technology offers new tools. AI-powered defense systems can observe network traffic, identify anomalies, and respond to threats faster than a human team. But the asymmetry of offensive and defensive uses remains. An attacker only needs to find one vulnerability; a defender must protect every entry point.
Taiwan as a testing ground
Taiwan has often been the first place where new cyber techniques are observed. Its dense digital infrastructure, combined with persistent political tension, makes it a natural laboratory for both attackers and security researchers. The July campaign fits that pattern. It is likely to influence how other governments think about AI-assisted threats.
The island's Ministry of Digital Affairs was established in 2022 to consolidate cybersecurity and digital policy. It has invested heavily in resilience, running regular drills and coordinating with critical infrastructure operators. The fact that a campaign of this speed could still make progress demonstrates how hard it is to defend against adaptive adversaries, even with strong preparation.
The ministry says it has since tightened monitoring and issued protective guidance across its agencies. That is a reasonable first step, but it may not be enough. If a handful of AI agents can rifle through government systems in four days, the next campaign is unlikely to wait politely for defenders to catch up.
The significance of speed
The compression of time is the part worth dwelling on. What used to be measured in the weeks a human crew needed to move laterally through a network can now be squeezed into a long weekend. This rewrites the arithmetic for everyone tasked with defending a network. Incident response teams that were trained to handle slow, methodical intrusions must now deal with attacks that move faster than they can report.
Speed also affects detection and attribution. When an attack completes in days rather than weeks, there is less time to observe and block the attacker. The data may be exfiltrated before defenders even realize a breach occurred. This places a premium on automated detection and rapid containment.
The Taiwan incident also underscores the importance of international cooperation. Cyber threats do not respect borders, and attackers often route their operations through multiple countries. Sharing threat intelligence across governments and private-sector firms is essential to staying ahead. The report from the cybersecurity firm and the subsequent confirmation by Taiwan's government show that such cooperation can work, even when political tensions complicate matters.
What this means for governments and businesses
For governments, the lesson is that AI-assisted attacks are no longer a distant possibility. They are happening now, and public-sector networks are prime targets because of the sensitive data they hold. Password management, multi-factor authentication, and network segmentation remain basic but critical defenses. Yet those measures are not enough on their own. Officials must assume that attackers will eventually get in, and plan for rapid response and recovery.
For businesses, the same principles apply. The tools that enabled the Taiwan campaign are available to anyone with a credit card. Companies that have not yet invested in AI-powered security or regular penetration testing are at risk. The barrier to entry for attackers is dropping, but the cost of defense remains high. That imbalance is likely to persist for some time.
The question of liability is also unresolved. If an AI agent causes damage, who is responsible? The developer of the agent? The operator who launched it? The platform that provided the underlying model? Similar questions are being asked in courts and legislatures around the world. The Taiwan case is a reminder that these are not abstract debates. The consequences of an autonomous hack can be measured in stolen credentials, leaked personnel files, and compromised national-security systems.
There is no conclusion to these developments. The campaign against Taiwan's government lasted four days, but its implications will unfold for much longer. Security teams in Taiwan and elsewhere will study the attack, update their defenses, and prepare for the next one. The AI agents, too, will improve. They will be harder to detect, faster to act, and more resistant to interference.
The only certainty is that the race between attackers and defenders is accelerating. Taiwan's brief but intense encounter with AI-assisted hacking is a warning to the rest of the world: tomorrow's threats may not wait for the humans to catch up.
Source: TNW | Security News