0
/ Pixabay

As artificial intelligence-driven cyberattacks targeting the financial sector proliferate, the virtual asset market has also entered the crosshairs of AI hacking. Going beyond using AI to discover vulnerabilities or refine phishing and social engineering attacks, cases have now emerged where attackers inject malicious prompts into AI agents to steal actual virtual assets. With cryptocurrency hacking also evolving to scale up attack speed and scope through AI, changes are required in the security posture of crypto exchanges.

According to blockchain analytics firm TRM Labs on the 6th, the index measuring the level of AI utilization in cryptocurrency crime stood at 54 out of 100 in 2026. This figure nearly doubled from 28 points in 2024 and represents an increase of approximately 40% compared to last year. Virtual asset hacks themselves have also surged. In the first half of this year, cryptocurrency hacks totaled 201 incidents, more than doubling from 83 incidents in the same period last year to set an all-time record.

Notably, the deployment of AI in hacking is viewed as having transitioned past the "initial stage" into an "expansion phase." TRM Labs analyzed that "rather than creating novel hacking techniques entirely on its own, AI is functioning as an 'amplifier' that allows attackers to carry out existing social engineering, credential theft, and vulnerability analysis attacks faster and at lower cost."

In fact, real-world cases of stealing virtual assets by attacking AI systems themselves have already occurred. In May, an attacker injected malicious instructions into the AI-based crypto trading service "Bankrbot," draining approximately 200,000 dollars' worth of virtual assets. According to foreign news reports, the attacker posted a transaction command encoded in Morse code on X and prompted the AI chatbot Grok to interpret it. When Grok relayed the decoded content to Bankrbot, Bankrbot interpreted it as a legitimate financial command and transferred roughly 300 million tokens to the attacker's wallet. It was revealed that no separate human verification process took place during this transaction.

As such AI-enabled hacking incidents grow, virtual asset exchanges have also begun leveraging AI as a defensive tool. Major global cryptocurrency exchange Bybit disclosed that it analyzed over 100,000 security alerts using AI in the first half of this year, uncovering high-risk vulnerabilities at a rate three to five times higher in AI-driven security audits than in manual reviews. It also explained that it blocked more than 30,000 suspicious withdrawal requests during the same timeframe, preventing potential user losses exceeding 700 million dollars. Furthermore, as attackers accelerate the speed of finding vulnerabilities with AI, exchanges are similarly using AI to automate code audits, penetration testing, and on-chain anomalous transaction detection.

However, concerns persist that AI agents could fundamentally alter the cost economics of hacking, accompanied by observations that deploying AI purely for defense is not enough. Ryan Kirkley, CEO of Global Settlement Network, warned at a blockchain symposium: "In the past, spending that much time to hack someone with 20,000 dollars was difficult, but now you can build an agent that attacks everyone," highlighting that attacks once confined to a handful of high-value accounts could expand via AI into large-scale automated operations.

Ultimately, AI security competition in the virtual asset industry is no longer just about "using AI to block AI," making it vital to reinforce multi-layered verification frameworks during asset transfer stages alongside AI-powered detection systems. An industry official remarked, "As the era of AI agents directly conducting financial activities draws nearer, a security structure where humans and systems exercise simultaneous oversight will become critical to ensure that erroneous judgments by AI do not lead to actual asset theft."

                                                                                                                Kim Min-ju