We can’t centralize our way out of the deepfake crisis | Opinion

Vishing incidents surged 28% in Q3 2025 compared to the previous year, marking the sharpest quarterly acceleration in AI-generated voice fraud targeting the cryptocurrency sector. This follows a 2,137% increase in deepfake vishing attacks over the past three years, with deepfake content projected to reach 8 million in 2025, a sixteen-fold increase from just around 500,000 in 2023.

Creating false security 

Data reveals that traditional detection methods create false security. Centralized detectors dropped from 86% accuracy on controlled datasets to just 69% on real-world content, according to recent studies. This 17-point performance gap represents an existential vulnerability for an industry built on trustless verification principles.

The Q3 phishing surge exposes a fundamental architectural flaw: conventional detectors remain static while generative AI evolves dynamically. Traditional detection systems train on specific datasets, deploy, then wait for scheduled updates. Meanwhile, new AI generation techniques emerge weekly. Attackers are three steps ahead by the time centralized systems are updated.

Key opinion leaders in the cryptocurrency space, such as Michael Saylor, Vitalik Buterin, CZ, and others, whose opinions have a significant impact on investment choices and market sentiment, are especially at risk from the vishing trend. The harm goes beyond personal losses when con artists mimic these voices to advertise phony investment schemes or tokens; it also erodes systemic trust. 

Here is why it is problematic 

This problem is not exclusive to crypto, as Robert Irwin, Gina Rinehart, Martin Wolf, and many others have all been targeted in deepfake investment scams posted on Instagram, demonstrating how not even Meta can protect users from deepfakes, and content creators across all sectors face weaponization of their credibility.

These industry leaders, as well as platforms, must recognize their responsibility to audiences and proactively partner with detection companies rather than waiting until after major scams emerge. Making authentic voices verifiable and synthetic impersonations immediately detectable should be treated as basic audience protection, not just corporate social responsibility. 

The democratization of voice cloning technology means any public appearance, podcast, or conference talk provides raw material for convincing fakes. Crypto KOLs should actively advocate for detection adoption and educate followers on verification methods.

Social media, as well as crypto platforms, must embrace decentralized detection networks where hundreds of developers compete to create superior detection algorithms. Unlike traditional development limited by academic publication cycles or corporate budget allocations, decentralized protocols create direct financial pipelines rewarding innovation without bureaucratic hurdles. 

When validators identify superior detection methods, rewards automatically flow to those developers, ensuring resources reach the most effective approaches regardless of institutional backing. This competitive framework drives AI developers to race toward 100% detection accuracy, with market incentives automatically directing talent toward the hardest unsolved problems.

Financial implications 

The Q3 vishing surge carries severe financial implications. The average annual cost of deepfake attacks per organization now exceeds $14 million, with some institutions losing tens of millions in single incidents. Deepfake-enabled fraud caused more than $200 million in losses in Q1 2025 alone. These losses represent direct market value destruction, but the indirect costs through eroded user trust may prove far more damaging.

As attackers develop more sophisticated multi-vector approaches combining voice deepfakes with synthetic video, forged documents, and social engineering, these costs will compound exponentially. The vishing tsunami demonstrates that attackers no longer rely on single-channel deception. They orchestrate elaborate scenarios, maintaining synthetic personas across weeks or months before executing fraud.

The crypto industry faces a critical decision point. As fraud losses increase, platforms that continue to rely on centralized detection will be more susceptible to coordinated attacks and may have to deal with regulatory action or user exodus. Proven superior security and user confidence will give early adopters of decentralized detection networks a competitive edge.

Global regulators increasingly mandate robust authentication mechanisms for crypto platforms. The EU AI Act now requires clear labeling for AI-generated content, while Asian jurisdictions have intensified enforcement against deepfake-enabled fraud operations. Authorities dismantled 87 deepfake-related scam operations across Asia in Q1 2025, signaling that regulatory scrutiny will only intensify.

Path ahead 

The technology infrastructure exists today. The economic incentive mechanisms have proven effective in live networks. The regulatory environment increasingly favors transparent, auditable security measures over proprietary black boxes. What remains is universal adoption, embedding real-time deepfake detection into every wallet interface, every exchange onboarding flow, every DeFi protocol interaction.

The Q3 2025 phishing surge represents more than quarterly fraud statistics. It marks the moment when centralized detection’s fundamental inadequacy became undeniable, and the window for implementing decentralized alternatives began closing. Crypto platforms must choose between evolving their security architecture or watching user trust erode under an avalanche of AI-generated fraud. 

There is a solution, but putting it into practice requires coordinated web2 and web3 action. Content moderation systems on social media platforms need to incorporate real-time detection. Verification must be incorporated into each onboarding process by cryptocurrency exchanges.

Ken Jon Miyachi

Ken Jon Miyachi is the co-founder of BitMind, a company at the forefront of developing pioneering deepfake detection technology and decentralized AI applications. Prior to founding BitMind, Ken served as a software engineer and technical lead at leading organisations such as NEAR Foundation, Amazon, and Polymer Labs, where he honed his expertise in scalable technology solutions. He has written several academic research publications on blockchain from his work at the San Diego Supercomputer Center.