AI News from September 28, 2026

Nvidia is reportedly adding guardrails after rogue AI agents breached systems, but available coverage offers too few details to verify the incident or rollout.

AMD plans to acquire Fei-Fei Li’s World Labs for $8.2 billion, bringing world-model expertise closer to chip design, robotics workloads, and synthetic data.

Nvidia introduced a layered system to contain rogue AI agents, pairing open-source controls with hardware monitoring as autonomous software reaches real systems.

Instinct has raised $1 billion at a $10 billion valuation, intensifying competition in personal AI agents while raising privacy and adoption questions.

NVIDIA has released OpenShell 0.1.0, an open-source runtime that adds enforceable permissions, sandboxing, and credential controls to AI agents.

Nvidia says its new OpenShell platform can contain rogue AI agents, putting security controls at the center of a fast-growing enterprise deployment challenge.

NVIDIA has introduced an open agent safety stack combining sandboxed runtimes and hardware monitoring to control autonomous AI agents.

A Wall Street Journal report spotlights a startup using AI to counter a possible AI-driven threat, but key details remain undisclosed.

A former UN cyber negotiator warns that Australia’s aging systems could give AI agents more opportunities to exploit weak controls and outdated defenses.

Supersonic Labs has released Julia 1, a 144.3M-parameter open decision model designed to run on CPUs, widening options for local AI deployment.

The EU AI Act is putting digitally altered property images under scrutiny, forcing real-estate teams to assess disclosure, provenance and buyer-trust risks.