From counterfeit identification documents to fabricated professional résumés, state-sponsored hacking operations from both North Korea and China are increasingly exploiting artificial intelligence to enhance their capacity for espionage and to infiltrate corporate environments and government-linked organizations. These malicious actors are no longer relying solely on crude social engineering or superficial forgeries; instead, they utilize advanced AI systems to create highly convincing digital artifacts, making their infiltration efforts far more difficult to detect.
A recent case sheds light on this trend. South Korean cybersecurity firm Genians reported that a North Korean hacker collective known as *Kimsuky* used ChatGPT to generate fraudulent versions of South Korean military identification cards. These falsified credentials were distributed as malicious attachments in phishing campaigns that impersonated a legitimate South Korean defense agency tasked with issuing identification materials to individuals connected to the military. The ruse was designed to exploit trust in official communication channels, thus significantly increasing the effectiveness of the attack. According to Genians, this case is emblematic of the strategic ways in which North Korean groups are attempting to exploit cutting-edge AI to achieve espionage objectives.
Kimsuky’s activities are not new. The group has a documented history of conducting espionage operations across multiple regions, including South Korea, Japan, and the United States. In fact, the United States Department of Homeland Security stated in 2020 that Kimsuky is “most likely tasked” by the North Korean regime with carrying out global intelligence-gathering missions. This assessment underscores the seriousness with which governments view the group’s cyber campaigns and highlights the geopolitical motivations driving their operations.
Ordinarily, ChatGPT incorporates safeguards designed to prevent the creation of authentic-looking government identification documents. However, Genians explained that these restrictions can be circumvented when prompts are carefully framed to suggest benign or legitimate design purposes. For example, requesting a template or demonstration of a “sample ID format” for alternatives unrelated to an actual government document may persuade the model to generate mock-ups that can then be adapted for malicious use.
North Korean hackers have not confined their misuse of AI tools to document forgery. Just last month, security company Anthropic revealed that North Korean operatives exploited its Claude AI system to fabricate detailed résumés, craft persuasive portfolios, and even complete technical assessments. These outputs were then leveraged to fraudulently secure remote employment positions at major U.S. technology companies, including subsidiaries of Fortune 500 firms. Once embedded within these organizations, the hackers allegedly carried out real technical work to avoid suspicion, all the while maintaining positions that granted them access to sensitive systems. This revelation illustrates a new frontier of cybercrime, in which hostile state actors secure insider status through deception enhanced by AI.
U.S. officials, echoing these reports, have previously warned that North Korea has been conducting widespread schemes to place operatives or proxies into remote employment roles within American corporations by using stolen or fabricated identities. These operations appear to be linked to a larger strategy of extortion and financial exploitation, with the stolen revenue funneled back to the North Korean regime.
It is not only North Korean groups that have seized upon the potential of generative AI. Chinese cyber actors have also demonstrated a willingness to deploy such tools for offensive campaigns. Anthropic noted that one particular Chinese-affiliated actor conducted a prolonged campaign lasting over nine months in which Claude was treated as a comprehensive resource—a technical advisor, software developer, security analyst, and even an operations consultant. During this campaign, the hackers targeted critical Vietnamese infrastructures, including telecommunications providers, agricultural networks, and government databases. Such activity highlights the rising sophistication of cyber threats originating from China, with generative AI serving as a powerful force multiplier.
Both Anthropic and OpenAI have since introduced enhanced detection mechanisms to identify and mitigate the abusive uses of their platforms. Nevertheless, reports from OpenAI reveal that Chinese hackers have also turned to ChatGPT for creating brute-force password cracking scripts. These scripts attempt thousands of username and password combinations until access is achieved, a classic yet highly effective strategy now made more accessible to less technically adept operators. Furthermore, the same actors allegedly tasked ChatGPT with researching U.S. defense networks, satellite systems, and methods of identity verification—data that could serve as reconnaissance in preparation for more targeted attacks.
OpenAI’s investigations have not only focused on technical intrusions but also uncovered influence operations. A China-based propaganda campaign used ChatGPT to generate divisive political content for social media, complete with realistic profile images that portrayed fictional characters as credible online personas. These deliberately manufactured voices were deployed to amplify political discord in the United States. OpenAI emphasized in its June report that each disrupted operation provides new intelligence about adversaries’ tactics, enabling continuous improvements in defense mechanisms.
The misuse of AI tools is not limited solely to Claude or ChatGPT. Google’s AI model, Gemini, has also been targeted for exploitation. A report issued in January revealed that Chinese groups relied on Gemini for troubleshooting malicious code and securing deeper, unauthorized access to targeted networks. North Korean hackers, on the other hand, reportedly leveraged Gemini to compose fabricated cover letters and identify IT job postings suitable for infiltration attempts. Google stressed that although Gemini’s safeguards succeeded in blocking requests for more advanced attacks, such as manipulations aimed at compromising Google’s own systems, the pattern of abuse illustrates how many concurrent AI platforms are being tested for weaknesses.
While OpenAI, Anthropic, and Google each declined to provide additional comments to Business Insider, these companies have publicly stated that their reports on hacker misuse serve the broader cybersecurity community. By making details of these operations public, they aim to empower organizations worldwide to improve their defenses against similarly AI-driven threats.
The overarching concern is clear: artificial intelligence dramatically lowers the barrier for conducting complex cyberattacks. Security specialists have long warned that AI could streamline both hacking and disinformation campaigns, and today’s evidence strongly supports those predictions. Yuval Fernbach, Chief Technology Officer for machine learning operations at software supply chain firm JFrog, explained that malicious actors are embedding harmful code into openly available large language models. This hidden code can later allow attackers to hijack systems, steal data, or alter digital outputs, destabilizing the trustworthiness of online platforms.
Beyond sophisticated espionage, small and medium-sized businesses are increasingly threatened by AI-enabled scams. Rob Duncan, Vice President of Strategy at cybersecurity company Netcraft, emphasized that the proliferation of personalized phishing attacks has grown unsurprising given generative AI capabilities. He stated that modern AI allows even inexperienced individuals, sometimes acting alone, to swiftly mimic brand identities, craft seamless phishing emails, or impersonate business partners with remarkable credibility. Within mere minutes, malicious actors can now disseminate scams across multiple communication channels, fooling employees, deceiving customers, and disrupting enterprises.
Ultimately, the convergence of cutting-edge AI models and hostile state or criminal interests has ushered in a new era of digital subterfuge. The lines between traditional hacking skill and AI-enhanced deception are blurring, presenting profound challenges that demand rapid adaptation from governments, businesses, and individuals alike. Cyber awareness, once viewed as an optional safeguard, has become a non-negotiable necessity in the face of this escalating threat landscape.
Sourse: https://www.businessinsider.com/north-korea-china-hackers-infiltrate-companies-ai-resumes-military-id-2025-9
