Episode Details

Today on UpNext AI: OpenAI lays out how it says it is aligning safety, security, transparency, and provenance work with Europe’s evolving AI rules; Microsoft Research introduces Echoverse, a high-fidelity training setup for computer-use agents; and we look at a new safety paper plus three quick headlines on robotics, identity security, and the real autonomy of AI agents.

Covered in this episode:
- OpenAI’s new Europe governance post and its framing around the EU AI Act
- Microsoft Research’s Echoverse environments for training computer-use agents
- A research writeup on improving the security and safety of generative models
- Google DeepMind’s Gemini Robotics 2 push into the physical world
- Okta’s reported acquisition of Permiso for about $200 million
- A Financial Times look at how autonomous AI agents really are

Source links:
- https://openai.com/index/advancing-responsible-ai-across-europe
- https://www.microsoft.com/en-us/research/blog/echoverse-deep-evolving-environments-for-computer-use-agents/
- https://doi.org/10.1184/r1/33062357
- https://www.wired.com/story/google-gemini-can-control-humanoid-robots/
- https://techcrunch.com/2026/07/30/okta-buys-ai-security-startup-permiso-source-says-for-about-200m/
- https://www.ft.com/content/56c3e0f1-6d74-4406-932e-86a9bd69b9bb?syn-25a6b1a6=1