Cyber risk enters the age of autonomous AI attacks

A recent cyberattack on Taiwanese government systems has provided a chilling preview of the future of digital risk. Security researchers have detailed what they term a 'near-autonomous attack', in which suspected Chinese operatives used publicly available AI tools to compromise the systems of Taiwan's nuclear safety agency, seven energy companies, and other government entities. The intrusion was not conducted by human operators in the traditional sense, but by a swarm of AI sub-agents, each assigned to specific targets and techniques. This represents a step-change in the sophistication and speed of cyber threats, moving beyond human-driven campaigns to automated warfare.

Over a four-day period in July, these AI agents independently executed a multi-stage attack with terrifying efficiency. They began by mapping the entire government IT ecosystem from a single portal, autonomously discovering connected systems and unauthenticated APIs. The agents then used this knowledge to breach a department's internal portal, solving its CAPTCHA security tests with perfect accuracy and cracking 85 user accounts via password spraying. In total, the agents exfiltrated more than 2,500 personnel records and a host of sensitive credentials. Critically, the framework demonstrated the ability to 'self-correct' and engage in 'learning cycles', searching for new exploits to use against its targets.

The most alarming aspect of this incident is that the attack framework was built using open-source AI agents, Hermes and OpenClaw. This means the capability to orchestrate such an attack is no longer the exclusive preserve of highly sophisticated state-sponsored groups. The tools are publicly available, suggesting that this new class of threat will inevitably proliferate, becoming available to a much wider range of malicious actors. The attack also highlighted the vulnerability of supply chains, as the agents, having breached the initial government systems, immediately pivoted to scan and attack connected IT vendors.

For finance leaders, this development demands a fundamental re-evaluation of cyber risk. The potential for financial and operational damage from an attack that unfolds at machine speed, rather than human speed, is an order of magnitude greater. Traditional incident response plans and defences may be too slow and reactive to counter an intrusion that is discovered, exploited, and weaponised in a matter of hours. CFOs must work with their security leaders to understand how their organisation's posture holds up against this new paradigm. This includes stress-testing detection and response times, reassessing the adequacy of cyber-insurance coverage for attacks by AI agents, and reinforcing due diligence on the security of all third-party vendors in the supply chain.