Verity describes a private corpus of attacks used to test its gate and publishes a map of known weaknesses. Users can test the gate; newly found defeats are recorded publicly and paid.
A peer-reviewed survey analyzes fully homomorphic encryption for privacy-preserving machine learning, outlining design frameworks, model adaptations, and trade-offs, while noting practical deployment limits and open research challenges.
The paper introduces TRACE, a method that reconstructs unlearned training data from pre- and post-unlearning model snapshots, revealing privacy risks in approximate unlearning for LLMs and suggesting defenses against parameter-difference leakage.
Cohere announces a public beta of Confidential Computing on Model Vault, offering hardware-backed, verifiable protection for managed AI inference on sensitive data. The system uses remote attestation, end-to-end encryption, and a trusted environment on NVIDIA H100 GPUs via Microsoft Azure to protect data in use during inference.
Translated title. Original: ¿Democracia digital real? Cómo votar de forma 100% anónima y segura en tu municipio con PAN-RAP - DEV Community
PAN-RAP Municipio presents an open-source, cryptographically anonymous voting system for small municipalities, aiming to ensure vote secrecy while enabling independent verification of results. It uses Blind Schnorr signatures to decouple voter identity from ballots, provides a public audit script, and is licensed under AGPL-3.0-or-later; the project is in a functional prototype stage with a Python/FastAPI backend and a Java frontend.
Translated title. Original: NVIDIA Confidential Bare Metal: computación confidencial para IA
NVIDIA introduces Confidential Bare Metal for AI, enabling large model developers to run sensitive workloads on third-party, non-virtualized servers with hardware-backed security, while maintaining control over models and data. The offering, announced at GTC Berlin with Google Cloud and Cohere, complements existing Confidential VMs and Confidential Containers, and expands hardware-anchored protections across supported cloud providers and NVIDIA systems.
Translated title. Original: Anthropic sécurise son IA Claude avec la technologie crypto de Zcash — Actualités TradingView
Anthropic unveils FlashProve, a verifiable inference infrastructure for Claude using zk-SNARKs built on Zcash cryptography to improve AI transparency and authentication. The piece frames crypto tech as essential to securing future AI models, notes current GPU costs as a hurdle, and cites data providers like ICE Data Services and FactSet.
Anthropic has acquired Opaque Systems, a Berkeley-founded security-focused AI company already in production deployments. Opaque plans to wind down after joining Anthropic, continuing its work on verifiable AI and cryptographic evidence, with past funding of $24 million and customers including ServiceNow and Accenture.
Fortanix extends its Confidential AI to Kubernetes via Confidential Containers, enabling AI inference inside hardware-backed trusted execution environments. The update addresses data-in-use risks during model and input processing, supported by attestation, policy-based key release, and one-click deployment on compliant hardware and platforms.
The document describes a private inference setup that uses a hardware-isolated Trusted Execution Environment, with data protections in transit via HTTPS and possible end-to-end encryption depending on integration. It also mentions attestation for identity assurance and the use of response signatures to verify integrity when available.