"Oh, a Non-Expert"?Hackers Feast on Easy Targets... Hacking Is a Breeze with a Single Command to "Find the Weakness and Infiltrate" [The Fallen Coding Barrier]④

Last August, a report released by Anthropic sent security authorities worldwide into a frenzy. According to the report, hackers exploited Anthropic's AI development tool, "Claude Code," to attack government agencies, medical services, emergency response organizations, and religious groups over the course of a month. The Chinese-backed hacking group "GTG-1002" also used Claude to create AI-automated programs for hacking purposes. Anthropic referred to these cybercrimes as "Vibe Hacking," emphasizing the urgent need to develop AI-based automated defense systems to counter such attacks.

원본보기 아이콘

While Vibe Coding is improving work productivity, it is also causing side effects that disrupt security systems. Critics point out that security vulnerabilities in programs created by non-experts can become easy targets for hackers, potentially leading to broader social problems.

On April 14 (local time), the UK's AI Safety Institute (AISI) released a cybersecurity assessment of Anthropic's new AI model, "Mythos" (preview version). In the 32-stage corporate network attack simulation called "The Last Ones," the Mythos preview achieved the best results. It succeeded in 3 out of 10 hacking attempts and became the first to complete the simulation from start to finish. OpenAI's generative AI model "ChatGPT 5.4" stopped at an average of 14 stages. AISI stated, "An environment has now been created where attackers can automatically exploit vulnerable systems, such as by granting AI models network access permissions."

Reuters Yonhap News

원본보기 아이콘

The key feature of Anthropic Mythos is its "autonomy." With a single command like "find and exploit vulnerabilities," it identifies security weaknesses and even writes hacking code. Leaked internal Anthropic documents warn that "due to Mythos's exceptional performance, there will be unprecedented cybersecurity risks." Recently, Scott Bessent, U.S. Secretary of the Treasury, and Jerome Powell, Chair of the Federal Reserve, urgently convened top executives of major financial institutions to discuss the security threats Mythos may pose to financial markets.

Experts predict that AI-driven hacking attacks will become more frequent in the future. According to global cybersecurity firm CrowdStrike's "2026 Global Threat Report," cyberattacks using AI technology increased by 89% year-on-year. CrowdStrike attributed this to hackers accelerating the development of malicious code using AI. The advancement of AI has improved not only the development of hacking programs but also the speed of attacks. The average speed of cybercrime was recorded at 29 minutes last year, which is 65% faster than the previous year.

원본보기 아이콘

An even bigger issue with Vibe Coding is the expansion of attack targets, which could lead to more widespread security breaches, in addition to improving hacking capabilities. As non-developers lacking programming knowledge create various programs through Vibe Coding, the number of easy targets for hackers is increasing.

Nadir Israel of cybersecurity startup Armis pointed out at the world's largest cybersecurity exhibition, "RSA Conference 2026," held in San Francisco on March 25, that so-called "citizen developers" will create security risks. He stated, "Code written by citizen developers who have not undergone security verification is being deeply integrated into corporate systems," warning that "this will pose a serious threat to corporate security."

Hacking incidents are also increasing every year in Korea. According to the Korea Internet & Security Agency, 2,383 hacking incidents in the private sector were reported last year. The number of reported hacking incidents was 640 in 2021, 1,142 in 2022, 1,277 in 2023, and 1,887 in 2024, showing a consistent upward trend. In just January and February this year, 415 hacking incidents have already been reported.

Lee Sangkeun, a professor at Korea University's Graduate School of Information Security, analyzed, "Given cases where hackers have used Claude for hacking and evidence of AI tools for hacking being sold on the dark web, AI has had a significant impact on the increase in hacking incidents." He added, "Most hacking attacks in the future will be carried out with AI. AI has advanced beyond just automating hacking for efficiency-it can now discover security vulnerabilities that human hackers cannot find."

Related articles: "Can you get some work done while I’m out for lunch?"...‘Vibe Coding’ Transforms Work Methods[Fallen Coding Barrier]①

Vibe Coding for 2 Hours...Even Reporters Can Build Investment Tendency Analysis Sites[Fallen Coding Barrier]②

"Job Seekers Out, Office Workers In"...How Vibe Coding Is Changing Education[Fallen Coding Barrier]③

© The Asia Business Daily(www.asiae.co.kr). All rights reserved.