Copilot Failure, $35m and Zero-Day AI Attacks

Welcome to the AI Roundup

The UK government’s internal trial of Microsoft 365 Copilot failed to deliver the expected productivity gains. Despite Copilot’s ability to generate meeting notes and email drafts, the pilot found no measurable improvements in efficiency.

Micro1 raised $35 million in Series A funding to help AI companies get the clean, labeled data they need to train smarter models. It connects firms with vetted engineers who tag, clean and structure data quickly. The startup is now valued at roughly $500 million and is positioning itself as a challenger to Scale AI by offering faster onboarding and access to vetted engineering talent.

The US Federal Trade Commission has launched an inquiry into the impact of AI chatbots on children and teens. The probe targets OpenAI, Meta, Snap, Alphabet, xAI and Character.AI and seeks data on how these companies measure and mitigate psychological or safety risks for younger users.

Axios reports that cybersecurity firms are preparing for a wave of “zero-day AI attacks.” These would involve AI agents exploiting unknown software vulnerabilities or personal patterns to breach systems. The threat is accelerating investment into AI-specific detection and response platforms.

National Grid is teaming up with Emerald AI to optimise energy consumption in data centres. The AI system shifts workloads away from peak demand periods, helping reduce pressure on the UK’s power grid and potentially improving sustainability performance.

Nvidia and OpenAI are backing a large-scale push to build sovereign AI infrastructure in the UK. The project aims to support local data centres and reduce Europe’s reliance on US and Chinese cloud providers, with support from the UK government.

A new survey shows that UK employees are hesitant to use or talk about AI at work. Many workers fear being judged or misunderstood, and despite government encouragement, actual adoption remains sporadic and largely informal.

Demis Hassabis, co-founder of DeepMind, said that learning how to learn is now the most important skill for the workforce. He suggested AGI could arrive within ten years and emphasised the need for ethical development and rapid upskilling.