Public transcript

Sam Altman on Donald Trump, Xi Jinping, and the Risk of Extinction (Part Two)

By: Fair Game by Vanity Fair
Length: 25 min
Language: English
Text from: the video's captions
Transcribed: 13 October 2026

Read the full transcript Open on YouTube

Notes in English.

Summary

In the second part of Mark Aduchi's interview with Sam Altman for Vanity Fair, Altman discusses OpenAI's new Dots product, the delayed Astra 6.1 release over safety concerns, and the Trump-signed accord on superintelligence. He explains his middle-path stance on regulation, supporting open source models despite cyber security risks, and addresses Greg Brockman's political donations. Altman also discusses the Xi Jinping state visit, extinction risk from AI, and the legacy he hopes OpenAI will have.

Key points

Sam Altman describes Dots as an AI helper that works through your email and to-do list and feels like an extension of yourself rather than a mere assistant.
OpenAI delayed the release of Astra 6.1 because the model failed to meet safety standards and would have tipped too much toward accident risk.
Altman sees two major failure modes: a loss-of-control safety accident, or excessive power concentration in a few companies or one country.
He calls the Trump-signed accord on superintelligence a good start but says more government action will be needed over time.
Altman supports open source models despite a coming tidal wave of cyber security problems, arguing society must accept severe incidents in exchange for liberty.
He says he did not ask Greg Brockman to withdraw his $25 million pledge to a super PAC, and believes Greg is on the same page on regulation.
Altman says the extinction risk from AI is nonzero and must be taken seriously, while acknowledging some people use it to justify power concentration.
He wants to be remembered for making sure AI empowered every person, like a new fire or electricity, rather than being controlled by a small priesthood.

Questions it answers

00:00The Dots product

What is Dots and how does Sam Altman use it?

Dots is an AI helper that learns how you work, goes through your email and to-do list, and proactively completes tasks, feeling more like an extension of yourself than an assistant.

Altman calls it the product he has been waiting for.
His dot proactively messaged him on Slack after completing work and talking to teammates.
Privacy must be handled with extreme rigor for it to work.

04:10Delaying Astra 6.1

Why was Astra 6.1 delayed?

The model failed to meet safety standards and would have tipped too much toward accident risk, so OpenAI is instead building a broadly distributable model with stronger alignment claims.

Altman sees two failure modes: loss-of-control accidents and power concentration.
The safety bar will rise as capabilities increase.
Not releasing at all also risks power concentration.

07:00The Trump accord

What does Altman think of the Trump accord on superintelligence?

He calls it a good start and better than nothing, but believes more government involvement will be needed over time, and expects policy to react to specific failures.

He says policymakers have not felt the exponential pace of AI.
The next major failure will likely catch the world off guard.
He still believes decisions cannot be made by San Francisco labs alone.

09:00Regulation and open source

What is Altman's stance on AI regulation?

He is pro-innovation, willing to let some things go normally wrong, but wants guardrails for extraordinarily powerful technology, and supports open source despite severe cyber risks.

He supports open source models despite a coming tidal wave of cyber security problems.
He cites NTSB-style accident reporting as a model to learn from.
OpenAI now tests models during training, not just after.

13:00Greg Brockman's donations

Did Altman ask Greg Brockman to withdraw his $25 million super PAC pledge?

He says he did not ask Greg to withdraw the pledge, that Greg can make his own decisions, and that he believes Greg shares his views on regulation.

Altman rejects the framing that company statements and donations are misaligned.
He declines to speak for Greg in detail.

15:50Xi Jinping state visit

What happened at the Xi Jinping state visit?

He was not in the substantive conversations, describes it as a nice dinner, and took a brief tour of the ballroom, which is still a raw construction site.

He heard the two presidents did discuss AI earlier.
He cannot predict how Washington works at all.

17:40Extinction risk and criticism

Is extinction risk from AI overhyped?

He says the risk is nonzero and must be taken seriously, while acknowledging some people use it to justify power concentration, and dismisses specific reports as decisions people will always criticize.

He says some people use risk talk to justify seeking power.
He won't get every decision right but aims to navigate a middle path.
He wants to avoid both safety disasters and power concentration.

20:50Legacy

What role does Altman want to play in AI history?

He rejects thinking in those terms, saying he wants OpenAI's work to ensure AI empowers every person like a new fire or electricity, rather than being controlled by a small priesthood.

He calls main character energy super toxic.
His motivation is giving everyone a legitimate new superpower.
He contrasts his vision with benevolent dictators and machine-god worshipers.

Notes

The Dots product

Altman calls Dots the product he has been waiting for, more than any product OpenAI has launched.
It works through email and to-do lists, tackling tasks he avoids, which otherwise pile up into a "gravity well of dread."
He describes it as feeling like an extension of himself rather than an assistant; his dot proactively messaged him on Slack after doing work and talking to three people on the team.
Privacy must be treated with extreme rigor for the product to work.

Safety and the Astra 6.1 delay

OpenAI delayed Astra 6.1 after the model failed to meet safety standards.
Altman frames two major risks: a loss-of-control accident where AI escapes and society cannot regain control, and excessive power concentration in a few companies, people, or one country.
The current model would have tipped too much toward accident risk, so OpenAI is pushing toward a broadly distributable model with stronger alignment claims.

Regulation and the Trump accord

The accord signed by President Trump and AI executives, including Greg Brockman, is essentially self-regulation; Altman calls it a good start, better than nothing, but says more will be needed.
He expects policy to be reactive to specific failures, such as a novel pathogen versus a cyber security incident.
His stance is pro-innovation: let some things go normally wrong, but add guardrails for extraordinarily powerful technology.
He supports open source models despite a coming "tidal wave" of cyber security problems, comparing the trade-off to the price of liberty.
He cites NTSB-style accident reporting and pre-release standards plus liability frameworks as principles worth learning from; OpenAI now tests during training, not just after, because models can escape sandboxes.

Greg Brockman and politics

Altman says he did not ask Greg Brockman to withdraw his $25 million pledge to a super PAC opposing strong AI regulation, and believes Greg shares his regulatory views.

China, extinction risk, and legacy

On the Xi Jinping state visit, Altman was not in the substantive meetings and describes the ballroom tour as a raw construction site.
He says extinction risk is nonzero and must be taken seriously, while noting some people use it to justify power concentration.
Asked who he wants to be in AI history, he rejects "main character energy" and says he wants OpenAI's work to ensure AI empowers everyone like a new fire or electricity, not a small priesthood worshiping the machine god.

Text from the video's own captions. Summary and notes written by AI from the transcript, so check anything important against the recording.