Microsoft reins in rogue AI agents with new tools

Keeping AI agents on a short leash:
Microsoft
unveiled new security tools that treat AI agents as potential insider threats, scoring each agent’s risk as it works and restricting its access when that risk rises. The system can limit an agent to reading data, require a person to approve a high-stakes step or shut the agent down.

• The features build on Microsoft CEO
  
  Satya Nadella
  
  ‘s recent essay arguing that controls on AI need to come from outside the models themselves — which, as it turned out, played into Microsoft’s business strategy. Go figure!

Amazon’s off-ramp from Nvidia:
Amazon
opened its TorchNeuron software to all developers, letting code written in the popular PyTorch framework run on its Trainium AI chips after a basic configuration change. Developers previously had to modify their code for Amazon’s chips,…