The most critical decisions in medicine, finance, and defense are moving from human deliberation to artificial intelligence. With that shift comes a new kind of vulnerability, one that software audits and compliance checklists cannot fully address. How can a hospital trust that the AI interpreting a cancer scan ran only the approved algorithm, on the approved hardware, and produced an unaltered result? Ziru Labs, a Cleveland-based startup founded in 2025, is betting the answer must be rooted in physics, not just policy.
Its platform, which it describes as a "physics-layer security for AI," is designed to enforce an AI workload's authorized behavior directly in silicon and generate cryptographic evidence of compliant execution [Perplexity Sonar Pro Brief]. The company's first deployment, a 15-patent-pending architecture called Project Phoenix, is targeting a minimum-working-prototype demonstration in the second half of 2026 [Ziru Labs, June 2026]. For founder and CEO Daniel Martin, the goal is to make high-stakes AI computation verifiable, not merely trusted.
The Hardware-Rooted Wedge
The company's wedge is a simple, if technically complex, argument. Software-based controls and logs can be manipulated. To provide incontrovertible proof of what occurred during an AI inference, Ziru argues you must anchor the authorization and the evidence in the hardware where the computation physically takes place [Perplexity Sonar Pro Brief]. This hardware-enforced trust layer is aimed at organizations where the cost of a rogue or compromised AI model is catastrophic.
Its identified target segments are a who's who of high-stakes environments: regulated enterprises like hospitals and banks, frontier AI labs, defense authorities, and sovereign entities [Perplexity Sonar Pro Brief]. The use cases are similarly critical, spanning medical-scan interpretation, autonomous systems, loan approvals, and national-security decisions. In each scenario, the need is not just for a secure AI, but for a provably secure one.
A Team Forged in Regulation
The company is emerging from stealth with a small, focused team. Founder Daniel Martin brings a background spanning decades of entrepreneurship in heavily regulated sectors, with cited experience involving SEC, FINRA, FDA, and HIPAA frameworks [Ziru Labs, June 2026]. Company materials state the broader founding team includes individuals with experience in U.S. intelligence-community and cryptographic-warfare roles, as well as ultra-low-latency systems engineering and production AI commercialization [Ziru Labs, June 2026]. This blend of regulatory, security, and performance expertise is a deliberate fit for the problem space.
While no public funding rounds have been announced, the company reports building a substantial intellectual property foundation, with a 23-invention portfolio and 16 provisional applications filed [Ziru Labs, June 2026]. Its public engagement includes a noted collaboration with Keeta Network, which posted in September 2026 about building toward agent payments with Ziru to prove an autonomous agent was authorized and remained within its mandate [Keeta, September 2026]. The company lists between one and ten employees [Ziru Labs, retrieved 2026].
| Metric | Value |
|---|---|
| Provisional Patents Filed | 16 applications |
| Invention Portfolio | 23 inventions |
| Target Prototype | H2 2026 timeline |
The Path to Proof
The ambition is vast, but the immediate path is defined by a series of technical and commercial milestones. The success of Project Phoenix hinges on demonstrating that its architecture can be implemented without crippling the performance or cost of the AI systems it aims to protect. Furthermore, it must navigate a complex ecosystem of chip manufacturers, cloud providers, and model developers to achieve adoption.
The competitive landscape for securing AI is crowded with software-focused players, but Ziru's hardware-rooted approach carves out a distinct, if narrower, lane. The primary counter-bet in the market is that robust software controls, continuous monitoring, and rigorous model governance are sufficient for even the most sensitive applications. Ziru's thesis is that for the highest tiers of risk, they are not.
Key execution risks for the young company are substantial:
- Technical integration. Embedding trust at the silicon level requires deep collaboration with hardware vendors, a sales cycle measured in years, not months.
- Performance tax. Any system that adds cryptographic proof and hardware checks must prove it does not impose a prohibitive latency or cost overhead on AI inference.
- Standardization. For the evidence Ziru generates to be universally accepted, it may need to become part of an industry or regulatory standard, a notoriously slow process.
For patients and clinicians, the disease state here is one of diagnostic uncertainty compounded by technological opacity. In oncology, for example, AI is increasingly used to identify patterns in radiology scans that may elude the human eye. The current standard of care involves a radiologist reviewing the images, potentially aided by a software tool that operates as a black box. There is often no verifiable chain of custody for the AI's decision, no cryptographic seal guaranteeing the algorithm wasn't tampered with between its certification and its use on a patient's scan. Ziru Labs is attempting to build that seal, arguing that when an AI's conclusion can alter a life, its provenance must be as solid as a surgeon's scalpel.
Sources
- [Ziru Labs, June 2026] Company Website | https://www.zirulabs.com/
- [Perplexity Sonar Pro Brief] Research Brief on Ziru Labs
- [Keeta, September 2026] Social Media Post on X | https://x.com/KeetaNetwork/status/2097757500963725641
- [LinkedIn, retrieved 2026] Ziru Labs Company Page | https://www.linkedin.com/company/zirulabs