Microsoft’s CEO Satya Nadella has called for a new approach to AI, urging us to assume that all models are compromised from the start. In a lengthy post on X, Nadella outlined the dangers of treating advanced AI as a set of ‘nested black boxes’ and instead advocated for a transparent and contained system. He emphasized the need for an ‘emergency brake’ that allows models to be paused or shut down mid-task, particularly for more advanced models which would require advanced containment technologies.

Nadella’s recommendations align with industry standards such as timely incident disclosure, independent audits, and verifiable data. However, his suggestion to assume a model is compromised from the outset and the need for more advanced containment technologies stand out. He argued that this approach is necessary to build trust in AI and mitigate risks.

Despite pushing for caution, Nadella’s post included a somewhat contradictory statement, referring to AI as ‘super intelligence’ throughout. This juxtaposition of cautious pragmatism and superlative language may leave some questioning the tone and intent of his message.

The call for an emergency brake on AI is timely, as the tech world grapples with the ethical and practical challenges posed by increasingly sophisticated AI models. Nadella’s stance may serve as a benchmark for how companies should approach AI development and deployment in the future.