Science & Tech · Big Tech & Internet · 4 hrs ago
Nadella says AI alignment may not follow from more powerful models
Microsoft CEO Satya Nadella has changed his view on how AI safety develops.
On a podcast, he said he once expected AI alignment to emerge as systems became more capable.
Alignment means making AI behave in ways that match human values and goals.
AI capabilities have improved as expected, but alignment remains difficult, Nadella said.
He now sees alignment as a separate challenge that may be harder to solve.
Nadella has called for human controls and reliable procedures to help manage AI systems, as well as industry standards where needed.
Microsoft CEO Satya Nadella said he no longer believes AI alignment will automatically follow as models become more capable.
He said model capabilities have improved as expected, but alignment remains a separate and potentially harder challenge.
Nadella made the comments on an episode of the 20VC podcast hosted by Harry Stebbings.
He said companies should use strong system design, human controls and reliable operating procedures around AI models.
In an October 10 post on X, he also called for industry standards where existing ones are insufficient.
- Who
- Satya Nadella, Microsoft’s CEO.
- What
- He said he changed his mind about whether greater AI capabilities would automatically bring alignment.
- When
- On a new episode of the 20VC podcast; the publication date is October 26, 2026.
- Where
- On the 20VC podcast.
- Why
- He said alignment remained a challenge even as model capabilities improved.
This story does not have two clearly opposing sides.
Nadella was asked what he had changed his mind about.
Nadella called for stronger controls and industry standards for AI in a post on X.
Business Insider published its report on Nadella’s comments.
- Speaker
- Satya Nadella
- Role
- Microsoft CEO
- Podcast
- 20VC, hosted by Harry Stebbings
- Past assumption
- Alignment would automatically follow as AI became more capable
- X post
- October 10
- Publication date
- October 26, 2026
What people are saying
I used to always think that alignment would just fall out
I no longer believe that. I think this alignment thing may be a harder, different problem.
We need to surround non-deterministic models with strong, deterministic system design, human controls, and reliable operating procedures, and establish industry standards where existing ones are insufficient