Data security in AI jumps 10 as privacy risks surge

Centralized AI faces growing privacy vulnerabilities

In traditional AI systems, storing data on third-party servers creates serious security gaps. Each request submitted can be centrally logged, leading to potential leaks of sensitive information. Recent high-profile cases—such as Samsung engineers experiencing code leaks and the redirection of South Korean user data to servers in China—highlight the reality of these risks.

These incidents demonstrate that when AI prompts are stored in accessible formats, commercially valuable or strategic business information may be at risk. Crypto market analyst Kaff emphasized this danger, noting that, “An agent’s system instructions are its alpha; if they can be read, they can be copied.”

An agent’s system instructions hold all its strategic value. If these instructions are transparently accessible, rivals could gain direct insight, paving the way for “MEV-like” advantages, explained Kaff.

Glossary: MEV (Maximum Extractable Value) refers to the highest profit that miners or validators can obtain by manipulating transaction ordering on blockchain networks.

Corporate studies paint a clear picture. According to McKinsey’s AI report for 2025, data security emerged as the top barrier for businesses, rising by 10 percent in just one year. In addition, 80 percent of organizations reported that AI-based applications led to unauthorized access to sensitive data.

Crypto-powered privacy solutions gain ground

Tech giants are developing their own security solutions. NVIDIA offers advanced security modes in its Blackwell GPUs; Apple processes user data via encrypted clouds. Across the industry, platforms from Meta to Google Cloud promote “confidential computing” features. However, all these options remain ultimately dependent on centralized frameworks.

In contrast, crypto-focused projects pledge transparency, censorship resistance, and verifiable infrastructure. The Venice platform, now with more than two million users, provides end-to-end encrypted memory and privacy. NEAR uses trusted execution environments (TEE) to process transactions without granting servers access to user data, even for day-to-day operations.

Nillion combines multi-party computation (MPC), holistic encryption, and TEE technology. The platform has stored over 640 million documents and handled more than 1.4 million transaction requests. Phala Network leverages Intel and NVIDIA-backed hardware to perform large language model (LLM) operations securely, nearly matching standard computational speeds.

Glossary: TEE (Trusted Execution Environment) is a hardware-based security technology that establishes an isolated area within the processor, ensuring data and code are processed securely against outside interference.

Gartner’s recent forecast projects that by 2029, secure computation environments such as TEE will be mandatory for 75 percent of untrusted infrastructure. This expectation signals a major opportunity for crypto-enabled privacy solutions to expand into the enterprise sector in the coming years.