NVIDIA's Huang Warns of AI Liability as VAST Launches Secure AI Platform

Key Takeaways

• <ul><li>VAST DataEnclave uses NVIDIA Confidential Computing to run Cohere, CrowdStrike, and Deepgram AI models in hardware-isolated, secure environments.</li><li>Nvidia CEO Jensen Huang warns AI firms could face civil or criminal liability for releasing products they cannot control.</li><li>Donald Trump and Xi Jinping met at the White House to discuss trade, AI, Taiwan, and global rivalry, with significant differences remaining.</li><li>Lattice Semiconductor reported record Q2 revenue of $201.1 million and launched the Mach-N2 FPGA family and Lattice Prompt AI design tool.</li><li>Abnormal AI launched a unified security suite covering governance, cloud security, employee guardrails, agent security, and threat investigation.</li><li>Nozomi Networks uses AI-driven visibility to unify IT, OT, and IoT monitoring for critical infrastructure protection.</li><li>Salmon introduced Execution Verification Infrastructure (EVI) to create verifiable records of AI agent execution.</li><li>Fastino Labs released GLiNER2.5-Decide, a 340M parameter open weight model for structured decision making that runs on CPU.</li><li>Companies without major tech resources are turning to creative financing like GPU collateral to build AI data centers.</li><li>VAST DataEnclave specifically targets regulated industries including financial services and healthcare for confidential AI workloads.</li></ul>

Cisco and VAST build secure infrastructure for confidential AI

Cisco is partnering with VAST Data to protect sensitive data while AI models process it. VAST DataEnclave runs AI workloads inside hardware-isolated environments. The system verifies hardware and firmware before releasing encryption keys. This keeps data, models, and infrastructure separate from unauthorized access. The approach treats confidential AI as a full-stack infrastructure challenge.

VAST DataEnclave secures AI models and sensitive enterprise data

VAST Data launched VAST DataEnclave, a confidential AI capability using NVIDIA Confidential Computing. It lets AI models from Cohere, CrowdStrike, and Deepgram run in secure environments. This protects both customer data and AI model intellectual property. The tool targets regulated industries like financial services and healthcare. It keeps sensitive data within trusted infrastructure while interacting with advanced AI models.

VAST DataEnclave launches to secure AI and enterprise data

VAST Data introduced VAST DataEnclave to securely integrate AI models with sensitive enterprise data. The platform uses NVIDIA Confidential Computing technology. It supports models from providers including Cohere, CrowdStrike, and Deepgram. The launch addresses data privacy challenges in regulated industries. It represents a step toward a more encrypted and secure AI ecosystem.

Abnormal AI launches unified security suite for enterprise AI

Abnormal AI released a unified security suite combining governance, cloud security, and threat investigation. The suite includes AI Governance, AI Cloud Security, AI Employee Guardrails, AI Agent Security, and AI Security Workbench. It uses behavioral AI to help security teams monitor AI tools and agents. The platform addresses both enabling AI use and defending against AI-related threats. It helps organizations govern AI activity based on actual behavior.

Nozomi Networks advances AI-driven critical infrastructure security

Nozomi Networks is improving security for critical infrastructure using AI-driven visibility. The platform unifies IT, OT, and IoT data into a single monitoring view. This helps organizations detect and respond to threats more quickly. Bachir Moussa explained that unified visibility is essential for continuous risk management. The approach reduces the risk of cyber threats, downtime, and data breaches in industrial environments.

BiCFlow-MER uses conditional flow to improve multimodal emotion recognition

BiCFlow-MER is a new framework for multimodal emotion recognition that combines discriminative and generative methods. It separates emotion cues from speaker style and word content to better handle conflicting signals. The model transports audio and text evidence into a structured emotion space using bidirectional flow. It outperforms existing methods on IEMOCAP, MELD, and the CASE benchmark.

Study outlines AI trading framework for European electricity markets

This paper proposes a formal architecture for AI supported trading in European electricity markets. It aligns AI systems with market coupling, cross border transfer limits, and regulations like REMIT, MiFID II, MiFIR, and EMIR. The framework includes decision state tracking, exposure accounting, and fail closed control logic. It also identifies where cross border coordination fails due to timing and permission differences.

Fastino releases GLiNER2.5-Decide, a 340M model that runs on CPU

Fastino Labs released GLiNER2.5-Decide, a 340M parameter open weight model for structured decision making. It takes text and a typed schema and returns answers with confidence scores and feasibility metadata. The model uses a DeBERTa-v3-large encoder and a constrained decoder that follows schema rules. It leads benchmarks on intent routing and runs with low latency on CPU and GPU.

Five takeaways from Trump Xi talks on trade, Taiwan, and AI

US President Donald Trump and Chinese President Xi Jinping met at the White House to discuss trade, artificial intelligence, Taiwan, and global rivalry. Trade tensions eased slightly but major disputes remain. Taiwan stayed a key security concern, and the two sides differ on AI visions. Xi warned against a US China confrontation, and the war in Iran was also discussed.

AI data centers turn to GPU collateral and new financing methods

Companies without the resources of major tech firms are using creative financing to build AI data centers. Methods include using GPUs as collateral and other novel funding structures. These approaches add risk if AI demand does not meet expectations. The trend is growing in the United States and Japan.

Nvidia CEO warns AI firms face lawsuits if models cause harm

Nvidia CEO Jensen Huang says AI companies should not release products they cannot control. He says they could face civil or criminal liability for damage caused by their technology. Huang made the comments in a New York Times interview. He also dismissed fears that AI will end the world as doomsday narratives.

Salmon launches tool to verify AI agent execution

Salmon introduced Execution Verification Infrastructure (EVI) to secure AI agents and autonomous systems. It captures execution as signed events and records state transitions. This creates a verifiable record of what happened during software operation. The tool is developed by Archipelo, Inc. and supports AI safety and governance systems.

Lattice Semiconductor rises on AI product interest

Lattice Semiconductor stock rose 4.8 percent amid investor interest in its AI and security products. The company launched the Mach-N2 FPGA family and Lattice Prompt, an AI-powered design tool. Lattice reported record second quarter revenue of 201.1 million dollars. The company guided toward a one billion dollar revenue run rate.

Sources

• Building a trusted foundation for confidential AI
• This Week In AI Chips - Secure AI Integration Advances with VAST DataEnclave Launch
• This Week In AI Chips - Secure AI Integration Advances with VAST DataEnclave Launch
• Abnormal AI brings governance, cloud security, and threat investigation into one suite
• Nozomi Networks advances AI-driven critical infrastructure security
• BiCFlow-MER: Orchestrating Discriminative and Generative Multimodal Emotion Recognition via Conditional Transport
• Functional Architecture of European Electricity Trading Markets: Requirements for AI Supported Trading Systems under Regulatory Constraints
• Fastino Releases GLiNER2.5-Decide: A 340M Open-Weight Decision Model That Runs on CPU
• 'Let's race, not wrestle': 5 takeaways as Trump-Xi navigate trade, Taiwan and AI
• AI data centers increasingly built on GPU collateral, other novel financing
• Nvidia CEO Says AI Companies Will Be Sued to Death If Their Frontier Models Keep Doing Horrible Things
• Salmon Introduces Execution Verification Infrastructure (EVI) for Securing AI Agents and Autonomous Systems
• Lattice Semiconductor Gains as Investors Appear to Revisit Recent AI and Security Product Launches