# Why Sam Altman and Greg Brockman Say Humans Must Stay in Control of AI

- Channel: [The New York Times](https://www.youtube.com/channel/UCqnbDFdCpuN8CMEg0VuEBqA)
- Published: 2026-11-18
- Source video: [Watch on YouTube](https://youtube.com/watch?v=J5EA7QeBm8k)
- Read on ReadTube: https://www.read.tube/p/videos/J5EA7QeBm8k
- Language: English

## Summary

**OpenAI Leaders Put Humans First**

Sam Altman and Greg Brockman argue that increasingly powerful AI must remain under deep human control rather than gradually taking over decisions about the future. They believe alignment is possible, but only if safety, monitoring, security, and governance advance alongside model capabilities. They defend iterative deployment and AI’s major benefits while acknowledging serious risks, calling for stronger shared standards and greater coordination across the industry. Their close partnership and OpenAI’s growing control over its technology stack are presented as important to navigating this transition responsibly.

Sam Altman and Greg Brockman say AI must remain under meaningful human control, rejecting the idea of a gradual handover of decision-making to machines. They argue that increasingly capable systems require stronger alignment, monitoring, and security work before capabilities advance further; Altman says pushing model intelligence without being able to verify its behavior would be irresponsible.

They believe alignment is technically achievable, but the required confidence must rise as models become more powerful. Brockman points to chain-of-thought monitoring as an existing safety tool whose reliability is declining as models solve more internally, creating a need for new monitoring methods and shared industry standards. Both describe safety during development—not just after deployment—as a major cultural shift for OpenAI and the wider industry.

OpenAI’s leaders defend iterative deployment despite serious security failures and criticism from former employees and regulators. Brockman says the company wishes the incident had not happened but believes confronting reality with successive, safer versions is less dangerous than unveiling a highly powerful system all at once. He describes a security sprint that redirected a quarter of production engineers toward defense and argues that frontier models must be used to protect systems against increasingly capable threat actors.

Altman and Brockman present AI as both a severe risk and a potentially transformative public benefit. They cite applications in chip design, advanced mathematics, cybersecurity, health information, and productivity, while emphasizing that commercial pressure and competition cannot override safety and security. They call for shared standards, greater transparency among laboratories, and societal debate about governance because companies cannot reliably judge one another’s safeguards in isolation.

Their partnership is portrayed as a central strength of OpenAI, built on rapid communication, trust, and prolonged debate over difficult strategic and technical questions. They say their differing views eventually converge or produce a deliberate disagreement, while the company’s future also depends on scaling compute, improving cost efficiency, and integrating its models, chips, data centers, consumer products, and enterprise services.
