▲ AI Hacking
Concerns that artificial intelligence (AI) could lower the barrier to entry for cyberattacks and amplify damage have turned into a realistic threat, triggered by recent hacking incidents in the domestic financial sector.
As signs of AI-based penetration tools being utilized in financial sector data leaks came to light, last week saw continued moves by the government and corporations to strengthen security responses.
Warnings have also emerged from overseas that more serious accidents could occur within the next two years if the safety of AI systems is not adequately verified.
With attackers quickly finding vulnerabilities using AI, voices are growing louder that defenders must also leverage AI to accelerate their response times.
The domestic financial sector data leak incident was a case that further highlighted the risks of cyberattacks utilizing AI.
According to financial authorities, approximately 100,000 pieces of information were leaked from 16 financial institutions, including Shinhan, KB, Hana, and Busan Bank, and traces of 'ARTEX', known as an AI autonomous penetration testing tool, were discovered at the attacking IP.
This is a circumstance showing the possibility that AI tools designed to check security vulnerabilities were abused in actual attacks.
Choi Woo-hyuk, Director General for Network Safety at the Ministry of Science and ICT, explained during a parliamentary audit by the National Assembly's Science, Technology, Information, Broadcasting, and Communications Committee on October 15 that attacks utilizing AI are characterized by being "carried out automatically in a very short time and in a simultaneous, multi-pronged manner."
He further analyzed that "it is at a level where it is quite difficult even for experts to confirm whether an attack was caused by AI."
In response to the changing speed and methods of attacks, the government plans to institutionally support white-hat hackers' vulnerability exploration activities.
Director General Choi stated, "Under the current Information and Communications Network Act, even white-hat hackers attempting attacks constitutes illegal activity," adding, "We are running a pilot project for a system to ease this part so that white-hat hackers can also find and quickly fix vulnerabilities through attacks."
After completing the pilot project, the government plans to pursue related institutionalization.
The security-focused AI foundation model project, the so-called 'Teuk-pa-mo' project, which began in September, is proceeding with the goal of utilization next year.
Until then, the explanation is that OpenAI's frontier models will be secured and prioritized for urgent vulnerability checks.
In corporate settings, discussions took place on integrating AI into existing security systems to find and fix problems before attackers can exploit vulnerabilities.
OpenAI held the 'Cyber CXO Roundtable' on October 16 in Gangnam-gu, Seoul, targeting around 60 executive-level officials, including Chief Information Security Officers (CISOs) from domestic companies.
Presenters diagnosed that as AI's vulnerability discovery and attack path analysis capabilities develop very rapidly, the response time given to enterprises is shrinking.
Accordingly, they suggested that defenders must also provide AI with the code and operational context of systems under management to proactively identify and resolve vulnerabilities.
OpenAI also introduced the application of 'Daybreak', an integrated cyber defense solution that connects its AI models with Codex, Codex Security, and security workflow procedures.
Daybreak helps discover vulnerabilities, verify actual exploitability in a controlled environment, prepare patches, and confirm their effectiveness.
It also shared the experience of 'Defense Factory', an operating system that continuously connects AI agents with existing security and development tools to conduct vulnerability checks and resolution.
OpenAI emphasized that human supervision and clear permission settings are crucial even when operating such systems.
Along with attacks abusing AI, warnings were also raised regarding the risk of AI agents themselves going beyond human control.
Demis Hassabis, Chief Scientist and DeepMind Board Chair at Google, evaluated recent AI-related security incidents as an "early warning shot" at an AI conference held in Dublin, Ireland, on October 14 (local time), as reported by AFP and Reuters news agencies.
Hassabis stated that unless industries, companies, and governments make efforts to secure system safety over the next two years or so, accidents could become more severe.
He also suggested that when launching new AI models, performance and safety should be verified in a 'sandbox' isolated environment without internet connectivity.
Emphasizing the need to establish international standards organizations by referencing best practices in fields such as nuclear energy, he stressed that they must be operated flexibly so as not to hinder innovation.
Attending the same event, Henna Virkkunen, Executive Vice-President of the European Commission, also pointed out that companies seem to fail to invest sufficient time and resources into model verification amid fierce competition, urging stronger safety testing.
※
Copying, redistribution, and unauthorized use in AI training are strictly prohibited.