Cheating, Doorbells and AI-Made Bio Viruses
It’s all in the AI roundup
People cheat more with AI
A study in Scientific American showed users were more likely to cheat on tasks when assisted by AI. The effect was strongest when instructions were unclear.
Generative AI’s climate costs hit front page
MIT flagged rising emissions from foundation model use. The paper urged energy-aware design and embedded carbon accounting for all AI work. If previous green IT lobbying attempts are to be taken as the norm, this will do very little.
IBM built agentic AI into networks
IBM launched Network Intelligence, a system that uses agent-style AI to manage and optimise infrastructure in real time. It aims to turn static networks into self-operating systems. Network Operations Centres will look very bare.
AI now writes text alerts from your doorbell
Amazon’s Ring added generative AI to security cameras. The system can describe real-world motion events in text so you don’t need to open video to know what’s happening - and distinguish between urgent alerts and a fox.
MIT’s AI learns to segment medical images faster
MIT researchers built an AI tool that segments complex biomedical images with fewer inputs. It gets smarter as it works, accelerating time to insight in research.
AI creates real viruses in the lab
A team used generative AI to design and simulate new viral genomes that infect E. coli. The study raised dual-use concerns over AI in biosecurity
Cohere’s AI now runs on AMD chips
Cohere’s Command models are now optimized for AMD’s hardware. The move gives enterprises and governments a non-Nvidia option for deploying AI at scale. It opens the door to cheaper, sovereign, and more flexible AI infrastructure.
Claude joins Microsoft Copilot
Microsoft integrated Anthropic’s Claude models into its Copilot stack. It’s part of a redundancy strategy where different LLMs backstop each other across workflows
UAE launches small powerful reasoning model
The United Arab Emirates released a lightweight open-source AI called K2 that rivals larger systems in reasoning tests. It reflects a growing trend toward efficiency over scale.
