AI Privacy Solutions Debut At Summit And Beyond

At the bustling AI Infra Summit 2025 in San Jose, a trio of pioneering companies—DESILO, Cornami, and JeongwooMaru—alongside tech giant Samsung, unveiled a new era for privacy in artificial intelligence. As the world’s appetite for AI-driven insights grows, so do concerns about the security of sensitive data. The latest breakthroughs promise to rewrite the rules, making it possible to harness the power of AI without sacrificing privacy—a feat that has long eluded the industry.

DESILO, a Korea-based privacy-enhancing technology (PET) startup, and Cornami, a leader in scalable computing architectures, made headlines with their joint announcement: the deployment of a fully homomorphic encryption (FHE)-based large language model (LLM). According to a press release published on September 10, 2025, their new solution processes sensitive data while it remains encrypted, achieving both speed and accuracy in real-world scenarios. This marks a significant stride in addressing the persistent privacy–performance tradeoff that has plagued sectors like healthcare, finance, and government.

For decades, enterprises faced a difficult choice—either encrypt data and endure sluggish AI performance, or process data in the clear and risk exposing private information. Traditional encryption methods, while secure, slowed computations to a crawl. Faster inference often meant lowering defenses. The Cornami–DESILO collaboration seeks to close this gap, enabling encrypted AI inference at speeds that approach those of plaintext processing.

Dr. Craig Gentry, Chief Scientist of Algorithms at Cornami—and widely regarded as the father of FHE—explained the technical leap that made this possible. “For decades, FHE was considered too slow for practical deployment. By accelerating encrypted computation, we are proving that enterprises no longer need to choose between privacy and performance.” He emphasized the importance of plaintext ciphertext matrix multiplication (PCMM), a process that impacts over 90% of LLM computation. “The value of PCMM is that it makes privacy-preserving AI practical. By enabling matrix multiplication—the core of LLM computation—to be performed efficiently and securely, we are closing the performance gap between fully encrypted and plaintext LLM inference, while also strengthening compliance, data sovereignty, and post-quantum security,” Dr. Gentry told attendees.

Combined with Cornami’s scalable Compute Fabric, this approach delivers performance that is orders-of-magnitude faster than previous encrypted methods. The implications are far-reaching. In healthcare, for instance, clinical trials generate massive amounts of sensitive data. Researchers need to analyze this information swiftly, but privacy regulations demand rigorous protection. With DESILO and Cornami’s solution, encrypted clinical trial data can be analyzed without ever exposing the underlying records—a breakthrough for compliance and patient trust.

The financial sector stands to benefit as well. Banks and investment firms, often hamstrung by strict data sovereignty and zero-trust requirements, can now leverage AI without compromising customer confidentiality. Government agencies and cloud service providers face similar challenges, and the new technology promises to streamline compliance across the board.

DESILO’s CEO, Seungmyung Lee, summed up the company’s philosophy: “Our principle has always been simple: never decrypt. With Cornami, that principle becomes practical. This collaboration lays the foundation for enterprises to safely unlock high-value data, and it aligns with the upcoming launch of our HARVEST™ platform, which will support global healthcare partners.” The HARVEST™ platform, set to debut in December 2025, aims to accelerate secure multi-party data collaboration, particularly in healthcare—a sector where privacy is paramount.

While DESILO and Cornami focus on keeping real data encrypted, JeongwooMaru, another innovative Korean startup, is tackling the privacy challenge from a different angle: synthetic data. Founded in September 2024, JeongwooMaru specializes in generating high-quality synthetic structured data for AI development, especially in industries with stringent data regulations. Their flagship product, RealDataEcho, produces synthetic data that boasts a 99.9% statistical similarity to real-world datasets—a claim certified by the Telecommunications Technology Association (TTA) and validated through official listing on AI Hub.

JeongwooMaru’s approach is rooted in experience. CEO Choi Euisoon, who spent 12 years managing financial risk and credit at Nonghyup Central Association, witnessed firsthand the limitations of using sensitive data in practice. “Structured data is more than just data that ‘looks plausible’. If the computational relationships between numbers or domain characteristics are lost, the data cannot be used for real-world AI training,” Choi explained. JeongwooMaru’s proprietary algorithms preserve these crucial relationships, ensuring that their synthetic data is not only privacy-safe but also useful for AI training and analysis.

The company’s on-premises solution allows organizations to customize data while meeting strict security requirements—a must for financial institutions, healthcare providers, and public agencies. With two patents registered for structured data synthesis and recognition from multiple government and industry bodies, JeongwooMaru is rapidly building trust in Korea’s data privacy landscape. Choi shared the company’s ambitions: “Our short-term goal is to stably supply synthetic data solutions to financial, public, and medical institutions in Korea, and to gain trust in the industry through demonstrated cases. Based on this, we plan to upgrade our solutions and prepare for overseas expansion by challenging support programs for technology-based startups.”

Meanwhile, Samsung is bringing privacy-enhancing AI directly to consumers through its Galaxy AI platform. On September 10, 2025, Samsung detailed a suite of privacy and control features designed to empower users in the age of intelligent mobile devices. Galaxy AI incorporates built-in safeguards that protect user data from the ground up, combining a hybrid approach of on-device and cloud-based AI. Key features like Live Translate, Interpreter, and Audio Eraser keep sensitive data securely on the device, while advanced settings offer users granular control over what data is processed online.

Samsung’s commitment is clear: personal data is never stored long-term or used for AI training, whether processed on-device or in the cloud. The Security and Privacy dashboard gives users visibility into which apps have accessed their data and allows them to manage permissions and sharing functions with ease. The Auto Blocker feature adds another layer of defense, automatically blocking potential threats and preventing unauthorized access. For those seeking even tighter control, Maximum Restrictions lets users block 2G service and avoid connections to less secure networks.

“Transparency and choice are the principles driving this work,” Samsung stated. The company’s vision is to put power back in the hands of users, letting them decide how their data is handled—whether shared or kept entirely on-device. With Samsung Knox protection and intuitive privacy settings, users can tailor their mobile experience without sacrificing security or usability.

As AI becomes ever more integrated into daily life, these advances signal a turning point. Enterprises and individuals alike no longer have to choose between privacy and performance. Whether through encrypted computation, synthetic data, or user-centric mobile safeguards, the future of AI is shaping up to be both smarter and safer—a development that’s sure to resonate with anyone who values both innovation and privacy.