AI News from November 6, 2026

OpenAI is working with Hacktron to uncover vulnerabilities, test third-party connections, and improve defenses as frontier AI systems expand.

OpenAI introduced BioBasicsBench, a 100-task test of AI agents on raw experimental data, giving builders a new lens on everyday wet-lab reliability.

Perplexity is reportedly positioning Teammates as AI agents for daily work, but launch details, pricing, performance and adoption remain unverified.

NVIDIA has released four open-weight Cosmos 3 models for interactive driving and robotics simulation, alongside tools for adapting and deploying them.

NVIDIA is demonstrating AI agents that assemble editable 3D simulations with Omniverse, linking design, robotics and industrial analysis workflows.

Thinking Machines Lab has launched Inkling 2 Small, while open weights remain pending, raising questions about access, benchmarks, and competition with Mistral.

The European Commission has selected a Capgemini-led consortium to assess safety and cyber risks in advanced AI models, signaling tighter scrutiny.

Airbnb plans to launch an AI travel agent in early 2027, combining trip planning with booking support and a visual interface built for travel inside its platform.

Palo Alto Networks is reportedly negotiating a $500 million Clutch acquisition, underscoring rising demand for security controls around AI agents and enterprise systems.

A Decoder report says OpenAI’s GPT-6 Astra underperformed on safety tasks, raising questions about covert refusals, training data, and oversight.

Dropbox is buying Tigris Data for $350 million to build DBX Agent Cloud, giving AI developers storage infrastructure while expanding Dropbox’s own AI tools.

Merge is using OpenAI’s GPT-6 Astra, ChatGPT and Codex across code review, marketing and sales work, offering a narrow view of enterprise AI adoption.