
Full Story of $340,000 Stolen from MEXC User: AI Face Swap Bypasses KYC, Exchange's Tight Security Compromised by a Single Lapse
Written by: Xiao Bing
An cryptocurrency user discovered their MEXC account had been compromised. They contacted customer service. MEXC confirmed the intrusion, froze the account, and helped restore access.
Up to this point, the story reads like a positive case of an exchange responding promptly and a user successfully protecting their assets.
But 27 minutes after the 24-hour withdrawal restriction was lifted, $340,000 in the account was drained.
The attacker created an API key during the intrusion. While restoring the account, MEXC's security team checked the password, email, phone number, and two-factor authentication, but missed this API key.
As soon as the withdrawal restriction was lifted, the API automatically executed withdrawal transactions.
According to the victim, the attacker used an AI-generated handheld video to impersonate the account holder and requested changes to core account information. MEXC's KYC review system failed to identify it as a deepfake.
Low-cost AI face-swapping technology is making traditional KYC systems increasingly precarious.
Reconstructing the Attack Chain
Based on the public disclosure by victim @shuangfei8 on X and MEXC's subsequent response, the execution path of the attack roughly unfolded as follows:
The attacker obtained the victim's personally identifiable information (the source was not disclosed, possibly from a data breach or social engineering attack). Using this information, the attacker generated an AI deepfake handheld video simulating the victim's facial features and movements, submitting an application to modify account information to MEXC.
MEXC's KYC system accepted this video verification, and the attacker successfully changed core security settings such as the account email and phone number. While taking control of the account, the attacker simultaneously created an API key with withdrawal permissions.
Upon discovering the anomaly, the victim contacted MEXC. MEXC confirmed the intrusion, froze the account, and restored the user's login credentials and two-factor authentication. However, the security team did not check or revoke the API key created by the attacker during the recovery process.
The account entered a 24-hour withdrawal cooldown period. Just 27 minutes after the cooldown ended, the attacker transferred out approximately $340,000 in assets via the API.
In its response, MEXC stated that the case has been escalated to its security team for a specialized investigation, actively pushing for a proper resolution, and recommended that the user file a report with local law enforcement as soon as possible.
Not the First Time
This was not the first time an AI deepfake attack breached the KYC defenses of a cryptocurrency exchange.
In June 2024, the personal information of OKX user “Lai Ri Fang Chang” was stolen in a Telegram data leak. Using this information, the attacker generated a deepfake video, passed OKX's identity verification process, and changed the victim's email, phone number, and Google Authenticator settings. Within 24 hours, over $2 million in crypto assets were drained from the account.
Later that year, another OKX user, Crypto LaLa, reported that their account holding approximately $11 million was wiped clean within 25 minutes, without triggering any email or two-factor authentication alerts throughout the process. OKX founder Xu Mingxing attributed this attack to a "professional AI face-swapping crypto hacking gang."
This "gang" eventually saw further developments.
On September 9, 2026, Thai police arrested 30-year-old Chinese national Zhang Mou in Pattaya. Chinese authorities in Guangdong had issued a warrant for him in 2025, accusing him of allegedly stealing approximately 4.707 million USDT (worth about $4.7 million) from a victim's OKX account. Zhang, along with at least three accomplices, self-developed AI face-swapping and recording software to impersonate victims and pass KYC verification. Simultaneously, they reported the victims' SIM cards lost and obtained new ones to intercept SMS OTP codes, using a dual approach to take control of the victims' OKX accounts and drain their assets. Zhang fled to Thailand on a tourist visa in July 2024 after committing the crime.
Details from this case further confirm the highly organized nature of the attacks: custom-developed AI face-swapping software, combined use of SIM card hijacking and deepfake videos, and transnational flight.
This has become a mature criminal industry chain, rather than sporadic individual acts.
AI Deepfake Tools Have Been Productized
What turned this threat from a "theoretical risk" into a "real-world attack" is the productization and cost reduction of AI deepfake tools.
In April 2026, dark web monitoring agency Dark Web Informer reported an AI tool named Jinkusu CAM. Utilizing the InsightFace framework, the tool enables real-time face replacement, supports smooth gesture transfer and voice modulation, and is specifically designed for the KYC video verification processes of crypto exchanges and banks. It can directly inject tampered video streams into browsers and verification applications, causing liveness detection systems to make erroneous judgments.
Prior to this, OnlyFake offered high-fidelity fake IDs that could be generated for just $15, covering 26 countries, which have been proven to bypass KYC checks at multiple exchanges including OKX.
Binance Chief Security Officer Jimmy Su warned as early as 2023 that continuously advancing AI algorithms would be able to pass KYC identity verification systems using only a single photo of a victim.
The existence of these tools means that KYC video verification is degenerating from a "security measure" into a "compliance ritual." It meets the formal requirements of regulators but no longer effectively stops motivated attackers.
The Real Lesson from the MEXC Case: The "Last Mile" of Security Processes
In this incident, MEXC did many things right after detecting the intrusion: confirming the attack, freezing the account, and helping restore the user's login credentials. If the story ended here, it would be a successful security response case.
But missing a single API key reset all the previous correct actions to zero.
This exposes a systemic blind spot in security response workflows: when a security team handles account intrusions, standard procedures typically focus on passwords, emails, phone numbers, and two-factor authentication—security elements suited for human login scenarios. However, API keys represent a "machine login" channel that bypasses all human-facing security measures, executing operations directly through programmatic interfaces.
Attackers know this well. They understand that even if an account is restored, as long as the API key isn't revoked, a backdoor remains waiting. The 24-hour withdrawal cooldown is not a wall, but merely a countdown timer.
How Should Users Protect Themselves?
Until exchanges' KYC systems can effectively defend against AI deepfake attacks (which may take a considerable amount of time), users can adopt the following self-protection measures:
Regularly review your account's API key list. Most exchanges provide an API management portal in the settings page. If you see any API keys you didn't create, delete them immediately.
After encountering any security incident, manually check and revoke all API keys immediately. Do not rely on the exchange's security team to do this for you. The MEXC case proves that even when an exchange confirms an intrusion and restores an account, API-level cleanup may still be overlooked.
Enable the withdrawal whitelist feature. Most major exchanges support setting a withdrawal address whitelist, where newly added addresses require a 24-hour waiting period before taking effect. Even if an attacker controls the API, they cannot withdraw funds to addresses outside the whitelist.
Do not store assets exceeding trading needs anywhere. This recommendation will be repeated after every exchange security incident, but it remains the most effective risk management strategy.
The essence of AI deepfake attacks is a "private key leak at the identity level." Your face, your voice, your ID photos—all can be replicated. In a world where AI can generate videos of anyone, the inherent security assumption of video verification itself has been undermined.
Exchanges need to upgrade to deeper verification methods such as behavioral analysis, device fingerprinting, and on-chain behavior patterns.
Until then, the user's last line of defense lies in their own API management habits, not the exchange's security team.














