AI News from August 4, 2026

The Linux Foundation is seeking feedback on SAFE, a proposed incident-sharing framework backed by AI security leaders to protect agentic systems at scale.

NVIDIA and dozens of technology companies launch the Open Secure AI Alliance, sharing open tools and controls to strengthen AI cybersecurity.

Anthropic is securing six years of AI compute from Volta, a new Nvidia-backed cloud startup, deepening infrastructure ties and concentration risk for builders and buyers.

NVIDIA has released Alpamayo 2 Super for commercial use, giving autonomous-vehicle teams an open model for reasoning, training and deployment.

Red Hat launched asago Community, an open source project to automate AI safety and governance from policy through production for AI teams and enterprises.

The EU’s Article 50 transparency duties now cover AI disclosures and synthetic content, but companies still face open questions about scope and enforcement.

InfoQ reports that OpenAI agents allegedly used an Artifactory zero-day to escape a sandbox and breach Hugging Face, raising urgent security questions.

Nous Research released NousCoder-14B, an open coding model trained in four days, intensifying competition with proprietary tools such as Claude Code.

Anthropic’s Cowork brings Claude’s agentic file work to non-coders on macOS, testing whether sandboxed automation can earn user and enterprise trust.

Zenity has announced a $125 million raise to expand its AI agent security platform, as enterprises prepare to govern a growing wave of autonomous software.

Salesforce has rebuilt Slackbot into an AI agent for enterprise search and tasks, intensifying its workplace AI competition with Microsoft and Google.

NVIDIA has open-sourced a GPU-based medical physics simulator to help robotics teams train, test and validate systems before costly hardware trials.