Modern Know Your Customer (KYC) systems have been marketed as improving trust in financial services. However, in reality, they have become one of the most fragile concepts in the industry. The greatest danger no longer comes from unknown hackers probing the perimeter, but from insiders and vendors who now reside equally within the system.
As KYC systems proliferate across banking, fintech and crypto platforms, access to the industry is still considered an acceptable cost of compliance. That level of tolerance is becoming increasingly indefensible, especially given that the work related to insiders has been counted almost 40 percent of cases by 2025.
At the same time, KYC workflows often require too much sensitive material—identity, biometric data and account details—to bypass cloud providers, authentication vendors and manual review teams. Each additional person, tool or system granted access extends the radius of the blast. The unfortunate truth is that most KYC stacks are designed in ways that make leaks not only impossible, but likely.
Recent breach data proves this. Nearly half of all incidents last year were attributed to two classic indicators of poorly designed KYC infrastructure: poor configuration and third-party vulnerabilities. The inconsistency alone had a measure 15 to 23 percent of all violations in 2025, while third-party disclosures contributed about 30 percent.
A specific example is last year’s breach of the “Tea” app, which was marketed as a platform focused on women. Passports and personal information disclosed after a The website is left publicly accessiblewhich shows how easily sensitive identity data can be leaked when basic infrastructure protections are not in place.
Exposure is no longer a theory
The measurement of vulnerability in centralized identity systems is now well documented. Last year he saw more than that 12,000 confirmed violationsresulting in the exposure of hundreds of millions of records. Supply chain breaches were devastating, with nearly a million records lost per incident.
These numbers are very important for KYC because the identity data is permanent. Compromised passwords can be reset, but passports, biometric templates and government-issued identifiers cannot. If KYC databases are copied, improperly controlled internally or accessed by vulnerable vendors, users may have to live with the consequences forever.
For financial institutions, the damage goes beyond the cost of responding to a breach. Erosion of trust directly affects entry, retention and regulatory processing, turning security failures into long-term commercial liabilities.
Financial services have not been maintained. Data from the Identity Theft Resource Center (ITRC) shows the volumes of breaches in that sector from a low of 269 cases by 2022 to over 730 in each subsequent year. This increase closely tracks the growing reliance on third-party compliance tools and outsourced review processes. Regulators may mandate KYC, but they do not require institutions to include sensitive data in ways that invite misuse.
Weak identity testing is a systemic risk
Recent legislative actions have emphasized how weak identity verification can be if it is treated as a box-ticking exercise. Lithuanian authorities dismantling of SIM-farm networks revealed the misuse of KYC controls and SMS-based verification We have built a formal communication infrastructure.
In that case, about 75,000 active SIMs were registered under false or recycled identities, allowing for massive fraud and account takeovers. The lesson is broad: when identity verification becomes a process rather than a dynamic, attackers adapt faster than controls can change.
AI-assisted compliance adds another layer of complexity. Many KYC providers—including platforms like Onfido and Sumsub—rely on centralized, cloud-managed AI models to review documents, flag confusion and score risk. In automated settings, sensitive inputs are transferred over direct control of the facility. Logs, information and training data may be stored under vendor policies rather than for regulatory purposes.
Security teams routinely warn employees not to upload confidential data to third-party AI tools. Yet many KYC programs do that exact behavior by design. When identity crosses organizational boundaries, insider abuse and vendor compromise become management issues rather than purely technical ones, an overview that offers little comfort to regulated businesses or affected users.
Re-creating the problem with secret AI
When systems take on trusted insiders and trusted vendors, a breach becomes a question of time rather than probability. The privacy challenges of AI start with a different assumption: sensitive data must remain secure even from those using the system. Cryptography enables this by executing code within isolated areas of hardware known as trusted execution areas (TEEs). Data remains encrypted not only at rest and in transit, but also during processing. Even administrators with root access cannot view its contents.
Research has shown that technologies such as Intel SGX, AMD SEV-SNP and remote proof can provide verifiable classification at the processor level. Used in KYC, secret AI allows identity checks, biometric matching and risk analysis to take place without disclosing raw documents or personal data to reviewers, vendors or cloud operators. Authentication can be cryptographically proven without copying sensitive files to shared databases. Internal access is changing from a matter of policy to a matter of physics.
Reducing internal visibility is not an invisible safety improvement. It changes who is at risk and reassures users that submitting identities does not need to be blindly trusted to invisible employees or subcontractors. Institutions reduce their credit history by reducing the access of blank documents to controlled data. Regulators are getting stronger assurances that compliance systems are compliant with data reduction principles rather than against them.
Critics argue that proprietary AI adds operational complexity or is dependent on hardware vendors. That concern should be looked into, but the complexity is there. It’s just hidden within obscure dealer stacks and manual review lines.
Hardware classification is readable in a way that human process controls are not. It is also consistent with a regulatory push toward demonstrative safeguards rather than mere policy guarantees.
A necessary change in KYC thinking
KYC will remain mandatory in all financial ecosystems, including crypto markets. What is not fixed is the property used to meet that obligation. Continuing to aggregate proprietary data and providing broad internal access is normalizing insider risk, a situation that is growing out of control given current patterns of breaches.
Secret AI does not remove all threats, nor does it remove the need for governance. However, it challenges the long-held assumption that sensitive data must be seen to be verified.
In an industry struggling to protect irrevocable personal information while maintaining public trust, that challenge is overdue. The next phase of KYC will not be judged by how much data centers collect, but by how little they disclose. Those who ignore internal risk will continue to pay for it. Those who redesign KYC around private computing will impose a higher level of compliance, security, and user trust, which regulators and customers may demand sooner than many expect.
