NVIDIA is highlighting new open models, applications, and tools designed to help developers build and run increasingly capable AI agents locally. The push emphasizes user control over deployment and customization across the company's local AI ecosystem.
NVIDIA has expanded its Nemotron family with Nemotron 3.5 Lightning, an efficiency-focused model for long-running agent workloads. NeMo Switchyard complements it with model-routing capabilities intended to improve the speed and cost efficiency of agentic AI deployments.
OpenAI unveiled GPT-5.6-Cyber, a specialized model for vulnerability research, penetration testing, incident response, and remediation. Built on GPT-5.6 Sol, it reduces refusals for some higher-risk exploit-development tasks and is limited to approved users.
CISA confirmed that ransomware operators are exploiting a high-severity Microsoft SharePoint remote-code-execution vulnerability. The flaw has been tracked as actively exploited since early July, raising the urgency for affected organizations to apply mitigations and patches.
Anthropic plans to add machine-readable watermarks to Claude-generated text and digitally signed provenance metadata to generated files where supported. The company says the transparency measures will also extend to outputs from older models.
AWS announced integrations that connect its Continuum AI security infrastructure directly with OpenAI Codex and Anthropic Claude Code. The move places security controls inside developers' AI-assisted coding environments and reflects AWS's strategy of serving as a cross-model security layer.