It was great to see you around, @alexgkendall
! I’m so impressed that the @wayve_ai autonomous vehicle navigated flawlessly in crowded Tokyo streets during rush hour, especially inside the shotengai (商店街) streets with lots of pedestrian traffic.
SAFETY
-
Wayve autonomous vehicle successfully navigates Tokyo rush hour streets
By
–
-
AI-Powered Robots Enhance Safety in Industrial Environments
By
–
My recent visit to Nokia Bell Labs left me deeply inspired. In the industrial robotics lab, I observed how AI is being developed to give robots an understanding of their physical environment enabling them to work alongside humans to enhance safety, efficiency, and
-
Did OpenAI Solve the AI Alignment Problem?
By
–
OpenAI vient-il de résoudre le problème de l’alignement des IA ? Ma dernière vidéo à ce sujet :
-
Calendar Invite Prompt Injection Data Exfiltration Attack Vector
By
–
Could this be abused as a prompt injection data exfiltration mechanism? Someone might trick the assistant into sending them a calendar invite where the description of the event includes private data pulled from other assistant-available sources
-
Moral responsibility for artificially created life
By
–
For a life that you did not create? A life you came to meet like every other person?
-
Monitor AI agents deployment tracking volume latency errors cost
By
–
Ship AI agents with confidence. See how to monitor deployed agents by tracking volume, latency, errors, and cost, applying AI judges for quality checks, and integrating user feedback https://
youtube.com/watch?v=RRpzbN
3hV3I
… -
Prompt Injection Attacks on LinkedIn: Vulnerability and Risk
By
–
It's delightful how easy it is to deploy working prompt injection attacks via LinkedIn
-
Frontier Safety Framework Updates for Advanced AI Systems
By
–
Important updates to our Frontier Safety Framework, with expanded risk domains for advanced AI and refined assessment protocols. https://t.co/Y1VlTjkgtn
— Demis Hassabis (@demishassabis) 23 septembre 2025Important updates to our Frontier Safety Framework, with expanded risk domains for advanced AI and refined assessment protocols.
-
Microsoft Tests Browser Actions for Copilot in Edge
By
–
Microsoft is testing Browser Actions on Copilot in preparation for its upcoming release of agentic browser features on Edge. "Copilot can browse the web and complete tasks using your Edge profile info."
-
AI Models Encrypt Reasoning Traces Via Previous Response IDs
By
–
I didn't quite realise this – the reasoning traces are actually encrypted and the model can access them via previous_response_id but you can't. Not that I particularly want to read them, but good to know.