Earendil is a hardware security startup that builds tamper response systems for AI compute infrastructure, including GPU clusters, to support hardware-enabled governance and compliance verification for AI development.
Earendil is a hardware security startup that builds tamper response systems for AI compute infrastructure, including GPU clusters, to support hardware-enabled governance and compliance verification for AI development.
People
Updated 03/19/26Funding Details
Updated 03/19/26- Annual Budget
- -
- Current Runway
- -
- Funding Goal
- -
- Funding Raised to Date
- $1,342,000
Org Details
Updated 03/19/26Earendil, LLC is a California-based hardware security startup founded on July 9, 2024 by Mehmet Sencan, a Caltech-trained applied physicist with over a decade of experience in chip and manufacturing technology. The company is headquartered in Tustin, California and develops hardware tamper response systems designed to provide physical security for AI compute infrastructure, particularly high-value GPU clusters. Earendil's primary product is SquibSink, a hardware tamper response system that provides instant, irreversible physical destruction of computing hardware upon tampering attempts. The system uses military-grade components and leverages existing industrial detonators positioned within modified GPU heatsinks to deny reverse engineering and secret extraction before adversaries can access critical components. SquibSink keeps unique device identifiers confidential and non-clonable, supporting export control compliance, gray-market prevention, and counterfeit protection. The system is designed for large-scale implementation across GPU arrays and custom hardware configurations, with applications spanning defense, aerospace, and proprietary technology sectors. Earendil's work is closely connected to the Flexible Hardware-Enabled Guarantees (flexHEG) ecosystem. FlexHEGs are mechanisms designed to enable verification and automated compliance guarantees that are multilateral, privacy-preserving, and trustworthy, for agreements regarding the development and use of advanced AI technology. The company received two grants from the Survival and Flourishing Fund's 2024 flexHEG round: $778,000 for its Security Layers Project (physical tamper-response hardware) and $564,000 for its Standards Infrastructure Project, totaling $1,342,000. Prior to founding Earendil, Sencan was brought on by Atlas Computing, an AI safety nonprofit, with support from the AI Safety Tactical Opportunities Fund (AISTOF), to advance tamper response mechanism components through technology readiness levels. His approach emphasizes using proven industrial components rather than custom-engineered destructive solutions, balancing effectiveness, cost, reliability, and implementation simplicity. Sencan's background includes a BS in Applied Physics from Caltech, where he worked in Professor Axel Scherer's laboratory on implantable electrochemical biosensors, and holds patents assigned to Caltech in that field. He previously served as COO of Integrated Medical Sensors and co-founded Wireless Diagnostic Systems. In August 2025, he presented at BSides Las Vegas on thwarting key extraction and supply chain attacks through physical destruction of GPUs. Earendil also maintains a technical blog covering topics including extending hardware security standards for AI development, privacy-preserving compute auditing, and technical implementation paths for safe AI hardware.
Theory of Change
Updated 03/19/26Earendil's theory of change is that effective governance of advanced AI development requires hardware-level enforcement mechanisms that cannot be bypassed through software alone. By building tamper-response technology that physically destroys computing hardware upon unauthorized access attempts, Earendil creates a foundational security layer for Flexible Hardware-Enabled Guarantees (flexHEGs). These mechanisms enable on-chip, privacy-preserving verification of compliance with AI development agreements, allowing international coordination on AI safety without requiring parties to expose proprietary information. The physical security layer ensures that software-based governance mechanisms embedded in AI accelerators remain trustworthy and cannot be circumvented, thereby enabling credible commitments between AI developers, regulators, and nations about the safe development and use of advanced AI technology. Earendil also works on the standards infrastructure needed to make these hardware guarantees interoperable and widely adoptable across different types of AI accelerators.
Grants Received– no grants recorded
Updated 03/19/26Projects
Updated 03/19/26Earendil develops hardware tamper response systems for AI compute infrastructure, enabling physical security layers that support governance and compliance verification of GPU clusters used in AI development.
Earendil develops standards and infrastructure for hardware-enabled AI compute governance, building the protocols and specifications that enable privacy-preserving verification of compliance with AI development agreements.
Discussion
No comments yet. Be the first to share your thoughts.