askST: AI tool Artex hijacked for hacking globally, drawing FBI warning; is Singapore vulnerable?
Sign up now: Get ST's newsletters delivered to your inbox
The artificial intelligence tool Artex has been hijacked by hackers to help them launch cyberattacks worldwide.
PHOTO: REUTERS
- The FBI says AI tool Artex has been “weaponised” for cyberattacks worldwide, automating hacking steps and helping attackers gain VIP access into victims’ systems in under three hours.
- Artex was made as an open-source security test tool, but hackers copied it and used it against banks, firms and government-linked targets in South Korea, Japan and other regions.
- Singapore has no confirmed Artex victims yet, but experts warn its connected economy is exposed. Organisations should tighten basics, check suppliers and use multi-factor authentication.
AI generated
SINGAPORE – It was supposed to help stress-test the cyberdefences of organisations. But the artificial intelligence tool Artex has been hijacked by hackers to help them launch cyberattacks worldwide, the US Federal Bureau of Investigation warned on Nov 13.
It said that Artex has been “weaponised to automate multiple phases of cyber intrusions at machine speed, from reconnaissance through data exfiltration”. In some cases, bad actors were able to gain VIP access to victims’ computer systems in under three hours.
Artex has been blamed for the wave of cyberattacks against South Korean banks and Japanese companies, as well as for a campaign that targeted hundreds of organisations globally in recent months.
What is Artex, is Singapore vulnerable, and how can organisations defend against it?
Cybersecurity experts answer these and more.
1. What is Artex?
It is a cybersecurity AI tool originally released on the code-sharing platform GitHub in July by Chinese cybersecurity engineer Li Puhua, who goes by the handle Autumn-27 on GitHub.
As an open-source tool, Artex was freely available for anyone to access, modify and use.
Li said Artex was intended to help organisations test their computer systems’ security by simulating attacks against them. But malicious actors have since used it for hacking.
After cybersecurity firms implicated Artex in a wave of cyberattacks against South Korean banks, Li said in October that it would become a closed-source project because of the “misuse of the tool”.
This means Artex would no longer be freely available.
However, various versions of Artex had already been copied and shared online for hackers to access.
Artex is not an AI model but a framework that connects different AI models – such as Claude, DeepSeek, GPT and Kimi – to cybersecurity testing tools used for simulated attacks, said Matthias Yeo, chief executive of CyberXCenter. This means Artex’s capabilities can evolve as the AI models it uses evolve.
2. How was Artex used in cyberattacks?
In earlier AI-assisted cyberattacks, hackers still directed most of the attack activities, such as instructing AI to help write malicious code, generate phishing e-mails and analyse stolen information, said Yeo.
But Artex takes over most of the work from hackers, requiring them only to type an initial instruction in natural, human-readable language for the tool to figure out and execute the next steps on its own without further supervision, and at machine speeds much faster than humans can manage.
This means Artex and similar tools help automate and speed up several stages of a cyberattack, letting attackers carry out familiar activities with less manual effort, said Gareth Cox, Horizon3 vice-president for Asia-Pacific and Japan. He added that AI has removed the constraint that attacks need skilled people to carry them out.
Said Shaun Williams, SentinelOne’s senior director for customer intelligence: “The key takeaway here is... the widespread availability of AI-powered tools like Artex allows threat actors to find and exploit vulnerabilities in internet-facing devices faster, more effectively, cheaper and with less skill than ever before. We aren’t seeing new or novel attacks.”
3. Who has been behind the Artex attacks?
Microsoft said that a China state-aligned group, Storm-0940, used Artex in September to automate attacks targeting hundreds of organisations globally across a broad range of industries. This included attacks that resulted in a bad actor gaining VIP access to an organisation’s cloud infrastructure in under three hours.
A China-based “targeted intrusion adversary” called Vapor Panda was identified by CrowdStrike as using Artex against more than 200 organisations globally, including foreign ministries in Latin America, Europe, Asia and the Middle East; semiconductor, defence and think-tank organisations in the US; and telecommunications providers in various regions.
Unidentified attackers also used Artex to hit seven South Korean financial institutions, resulting in the theft of data belonging to about 67,000 people. The earliest case emerged in late September.
Then in October, GMO Cybersecurity by Ierae said there were signs that Artex had been used in a spate of cyberattacks in Japan that affected around 80 companies.
4. Could Artex be behind the recent spate of cyberattacks globally?
Partially, going by the known cases. But other AI tools have been used and there are other factors.
CyberXCenter’s Yeo said the attack surface available to threat actors has increased massively.
“Organisations are increasingly reliant on... all sorts of inter-connected systems. That massively increases their attack surface.”
And the barrier to entry for attackers to target them has also been greatly reduced with AI.
This means that while organisations had time to find and patch vulnerabilities in the past, attackers are now using AI to find them far faster than organisations can, Yeo said.
While AI has helped defenders identify threats and fix flaws faster, attackers are gaining an edge, he added. This is because defenders need to take many things into consideration, such as data protection concerns and the risk of false positives and negatives with AI detection.
But attackers are not burdened by organisational constraints and can simply take advantage of a tool once it becomes available, he said.
The rise of AI tools like Artex makes the situation worse because they can carry out attacks involving multiple steps with minimal human involvement, Yeo added.
Open-source AI cybersecurity testing tool Cairn has also been paired with AI models to autonomously attack at least 19 organisations, including some in North America, Palo Alto Networks said on Nov 12.
5. Can we prevent open AI systems from being weaponised by hackers?
Complete prevention is unlikely, particularly once capable open systems have been released and downloaded, said Cox, adding that they can be modified and stripped of their safeguards.
“Developers cannot monitor or suspend these independent copies in the same way they can restrict access to their own platforms,” he said. This is unlike closed AI services where providers have greater control over how their systems are accessed and used, such as monitoring suspicious activity and suspending users to make malicious use more difficult.
He suggested that AI developers should build safeguards into their AI models before release, including assessing potentially dangerous capabilities and making safeguards tougher to bypass.
6. How susceptible is Singapore to Artex attacks?
There are no known victims of Artex attacks in Singapore yet, but the country needs to take the threat from Artex and similar tools seriously, said CyberXCenter’s Yeo.
He explained that Singapore has a highly connected and digitally dependent economy, with financial institutions, telecoms networks, government services and critical infrastructure relying heavily on interconnected digital systems. This makes organisations of all sizes here, not just large enterprises or government agencies, potential targets.
“Smaller suppliers and third-party service providers could become entry points into larger organisations,” said Yeo.
7. What can organisations do to protect themselves?
Many organisations are now thinking about using AI to fight AI by matching the speed of their defences to that of attacks.
But Yeo warned that while using AI defensively has benefits, it is still paramount for organisations to get the security fundamentals right, improve their ability to detect threats early, and ensure that people, processes and technology can respond at the speed of modern attacks.
“It does not matter whether you have the latest AI product if you are not aware of which of your systems are exposed to the internet,” said Yeo.
Scott Flower, Fortinet’s head of security intelligence for the Asia-Pacific region, said organisations should look beyond their main systems and review less visible parts of their environment, such as those used by their partners and suppliers, sales support tools and legacy services.
“Organisations should keep an accurate inventory, remove unnecessary access, use multi-factor authentication and limit permissions to what each account or service needs,” he said.
