Microsoft CEO Satya Nadella called for AI models to have an emergency brake, emphasizing the need for greater control and transparency in AI systems. Nadella’s comments come amid growing concerns about AI safety and incidents where companies have reportedly lost control of their models.
Nadella outlined a framework for AI safety, including the separation of models from the systems that manage them, and the externalization of controls and safeguards. He also stressed the importance of documenting every meaningful model action with tamper-proof evidence.
"We must assume a model is compromised and contain it from the start," said Nadella. "Think of it like an emergency brake." This approach aims to ensure that authorized personnel can pause or shut down a model mid-task, enhancing accountability and reducing risks.
The announcement follows Anthropic CEO Dario Amodei’s plan for more cautious AI development, highlighting a broader industry push for improved safety measures. Nadella’s remarks reflect a growing recognition of the need for transparency and control in AI systems.
Microsoft did not specify how the emergency brake would be implemented, and the company raised questions about the effectiveness of current safety measures. The next steps involve further discussions on AI governance and safety protocols.
Source: techcrunch
