Participated · Seed · Aug 2026

Attestable develops zero-knowledge proof technology intended to verify which calculations an AI system performs and what data it uses while keeping models and sensitive information confidential. The founders—CEO Yogev Bar-On, CTO Shahar Papini and VP of R&D Shahar Samocha—are mathematicians with backgrounds including Unit 8200, StarkWare and Safe Superintelligence. The company says it has achieved a mathematical breakthrough enabling rapidly verifiable proofs applicable to AI labs and critical organizational uses. Attestable’s stated longer-term ambition is to become part of the infrastructure and regulatory fabric for advanced AI systems used by labs, governments and large organizations. It currently employs about 15 people, primarily in Israel with a small team in California, and reports relatively low infrastructure costs due to its mathematically driven work. The company raised $20 million in November 2025 to grow the team and advance its technology.

Participated · Seed · Aug 2026

Corma builds AI models trained for defensive cybersecurity rather than offensive capabilities, focusing on tasks like analyzing logs and audits and maintaining consistent responses across many actions. The company deployed its first model six weeks before the announcement to Fortune 100 and Fortune 500 customers across healthcare, financial services, energy, critical infrastructure, and retail. Corma says its model has reduced threat response times by 94% in adopting organizations. Founded in 2025 and based in Tel Aviv and San Francisco, the startup is emerging from stealth with seed funding. With the $60 million seed round, Corma intends to expand the data and training that power its models and grow its engineering and research teams. Leadership positions and investors have highlighted the company’s combination of defensive security expertise and frontier AI research as a core strength.

Participated · Seed · Sep 2025

MetalBear is the creator of mirrord, an open-source Kubernetes development solution that proxies local code into remote production-like cloud environments with a single IDE toggle. By intercepting I/O at a low level, mirrord enables developers to test microservice changes against live databases, APIs, queues, and third-party services without deploying anything. The enterprise edition adds traffic routing and queue splitting so multiple engineers can safely share one staging cluster, eliminating costly per-developer environments. The product is already used by thousands of developers at companies such as NVIDIA, AWS, and Apple, with early adopters reporting 80 % faster test iterations and 30 % fewer production bugs. MetalBear employs 25 people across 14 countries and offers mirrord under an open-source model with a paid enterprise tier. As AI speeds up code generation, the company positions mirrord as the critical layer that removes the new testing and integration bottleneck in cloud-native development.