GEGGL

Microsoft Corp. (NASDAQ:MSFT) CEO Satya Nadella said he no longer believes artificial intelligence alignment will emerge automatically as AI models become more powerful, warning that ensuring AI systems behave safely could be a harder challenge than he previously expected.

Nadella Reverses Course on AI Alignment

On Monday, speaking on the 20VC podcast hosted by Harry Stebbings, Nadella reflected on how his views on AI development had changed over the past year, pointing to the growing challenge of AI alignment.

Nadella said he had previously believed alignment would emerge naturally as AI capabilities improved.

"I used to always think that, you know, alignment will just fall out. I no longer believe that," Nadella said. "So I think this alignment thing may be a harder, different problem."

Nadella Rethinks AI Scaling Assumptions

He also discussed his earlier belief in AI scaling laws, which suggest that increasing computing power, training data and model size can improve AI capabilities.

He recalled sharing those beliefs around 2018 and 2019, when prominent AI figures, including OpenAI CEO Sam Altman, Space Exploration Technologies Corp. (NASDAQ:SPCX) Elon Musk and Ilya Sutskever, were associated with the push to scale AI systems.

AI Evaluations Could Become Standard Practice

Asked what might seem unusual today but become commonplace within five years, Nadella pointed to AI evaluations, or "evals."

These tests and benchmarks help assess AI systems’ performance, reliability and potential weaknesses before deployment.

Nadella Warned of AI Risks

Earlier this month, Nadella urged companies to treat AI models as potential insider risks, warning that advanced models could be compromised.

He said companies struggled to trace model behavior to specific training data or weights despite giving AI systems access to sensitive data and critical actions.

Nadella called for "strong, deterministic system design, human controls, and reliable operating procedures."

AI Safety and Human Oversight Concerns

Last week, Anthropic warned that AI safety remained unresolved, citing cybersecurity risks and gaps in detecting dangerous behavior.

It called for stronger oversight and federal standards, while committing $350 million to research labor-market disruptions and support early-career workers.

Previously, Altman said he was "very uncomfortable" with treating AI as a deity, warning that surrendering human judgment to AI models was "a real safety issue."

Disclaimer: This content was partially produced with the help of AI tools and was reviewed and published by Benzinga editors.

Photo courtesy: Shutterstock