Skip to content

Simulation results · 13 September 2026

Oversight yes, slowdown no: simulating the reaction to "We Must Pace the Frontier"

We put "We Must Pace the Frontier" to 4,491 simulated people in three societies: Tech Twitter (1,884), US tech leaders (1,526) and the US public (1,081). They back the oversight in the plan and reject the slowing down. Independent evaluators inside the labs get 81–86% support in every group; slowing development substantially gets 12–28%; and only 5–9% would let a lab decide for itself when it is safe to keep going.

How were these results produced?

At Artificial Societies we build populations of AI personas grounded in the public profiles of real people and survey them the way a pollster would survey real people. Each of the three societies here answered the same ten multiple-choice questions about the specific proposals in Amodei's essay, with a one-line reason. The numbers should be treated as a forecast of how each group will react, not as a poll of real humans. Roles shown in the responses are kept only where they are generic job titles, so no individual is identifiable.

Read the essay: We Must Pace the Frontier

Question 1

How fast should AI development go?

“When advanced AI systems may create serious societal risks, which approach to further capability development comes closest to your own view?”

Tech Twitter n=1,884 · US Tech Elite n=1,526 · US General Population n=1,081

  1. Slow development substantially until safety work catches up

    Tech Twitter 12% · US Tech Elite 17% · US General Population 28%

  2. Continue developing, but only with stronger safety safeguards

    Tech Twitter 31% · US Tech Elite 42% · US General Population 39%

  3. Continue at roughly the current pace while improving safeguards

    Tech Twitter 27% · US Tech Elite 26% · US General Population 21%

  4. Accelerate development because the benefits outweigh the risks

    Tech Twitter 19% · US Tech Elite 7% · US General Population 5%

  5. Prioritize staying ahead of international competitors over slowing development

    Tech Twitter 11% · US Tech Elite 9% · US General Population 7%

Question 3

Should labs be required to host independent evaluators?

“Do you support or oppose requiring frontier AI companies to give independent external evaluators ongoing access to assess safety practices and report serious incidents?”

Tech Twitter n=1,884 · US Tech Elite n=1,526 · US General Population n=1,081

  1. Strongly support

    Tech Twitter 41% · US Tech Elite 43% · US General Population 53%

  2. Somewhat support

    Tech Twitter 40% · US Tech Elite 40% · US General Population 33%

  3. Neither support nor oppose

    Tech Twitter 9% · US Tech Elite 8% · US General Population 9%

  4. Somewhat oppose

    Tech Twitter 7% · US Tech Elite 6% · US General Population 4%

  5. Strongly oppose

    Tech Twitter 2% · US Tech Elite 3% · US General Population 1%

Question 10

Who should decide when it is safe to keep advancing?

“Who should have the primary authority to decide whether safety evidence is strong enough for a frontier AI company to make further capability advances?”

Tech Twitter n=1,884 · US Tech Elite n=1,526 · US General Population n=1,081

  1. The company's own leadership and safety teams

    Tech Twitter 9% · US Tech Elite 8% · US General Population 5%

  2. Independent external evaluators

    Tech Twitter 18% · US Tech Elite 15% · US General Population 12%

  3. Government regulators

    Tech Twitter 6% · US Tech Elite 10% · US General Population 20%

  4. A joint process involving companies, independent evaluators, and government regulators

    Tech Twitter 53% · US Tech Elite 57% · US General Population 55%

  5. An international body with participating countries

    Tech Twitter 3% · US Tech Elite 5% · US General Population 7%

  6. No one should have authority to limit further advances

    Tech Twitter 11% · US Tech Elite 5% · US General Population 2%

Question 6

Should governments limit advances when safety evidence lags?

“Do you support or oppose government-backed limits on further frontier AI capability advances when safety evidence has not kept pace?”

Tech Twitter n=1,884 · US Tech Elite n=1,526 · US General Population n=1,081

  1. Strongly support

    Tech Twitter 22% · US Tech Elite 36% · US General Population 48%

  2. Somewhat support

    Tech Twitter 44% · US Tech Elite 43% · US General Population 34%

  3. Neither support nor oppose

    Tech Twitter 14% · US Tech Elite 10% · US General Population 9%

  4. Somewhat oppose

    Tech Twitter 14% · US Tech Elite 8% · US General Population 6%

  5. Strongly oppose

    Tech Twitter 5% · US Tech Elite 4% · US General Population 3%

Question 2

How do people react to the proposal overall?

“Overall, how do you react to Amodei's proposal to pace frontier AI development?”

Tech Twitter n=1,884 · US Tech Elite n=1,526 · US General Population n=1,081

  1. Strongly support it

    Tech Twitter 19% · US Tech Elite 27% · US General Population 36%

  2. Somewhat support it

    Tech Twitter 36% · US Tech Elite 40% · US General Population 39%

  3. Feel mixed about it

    Tech Twitter 25% · US Tech Elite 19% · US General Population 17%

  4. Somewhat oppose it

    Tech Twitter 13% · US Tech Elite 9% · US General Population 6%

  5. Strongly oppose it

    Tech Twitter 7% · US Tech Elite 5% · US General Population 3%

Question 4

Would embedded evaluators stay independent?

“How likely is it that external evaluators with employee-like access inside frontier AI companies would remain independent in their oversight?”

Tech Twitter n=1,884 · US Tech Elite n=1,526 · US General Population n=1,081

  1. Very likely to remain independent

    Tech Twitter 8% · US Tech Elite 15% · US General Population 13%

  2. Somewhat likely to remain independent

    Tech Twitter 30% · US Tech Elite 43% · US General Population 39%

  3. Equally likely to be independent or influenced

    Tech Twitter 30% · US Tech Elite 23% · US General Population 29%

  4. Somewhat likely to be influenced

    Tech Twitter 26% · US Tech Elite 15% · US General Population 16%

  5. Very likely to be influenced

    Tech Twitter 6% · US Tech Elite 4% · US General Population 4%

Question 9

What is the biggest concern about the proposal?

“Which concern, if any, is most important to you when considering Amodei's proposal?”

Tech Twitter n=1,884 · US Tech Elite n=1,526 · US General Population n=1,081

  1. AI systems causing large-scale security or safety harm

    Tech Twitter 25% · US Tech Elite 44% · US General Population 47%

  2. AI development becoming too concentrated in a few companies

    Tech Twitter 22% · US Tech Elite 10% · US General Population 12%

  3. Government rules slowing useful innovation

    Tech Twitter 19% · US Tech Elite 11% · US General Population 7%

  4. The US losing its AI advantage to China

    Tech Twitter 20% · US Tech Elite 21% · US General Population 12%

  5. Independent oversight becoming ineffective or captured

    Tech Twitter 12% · US Tech Elite 10% · US General Population 17%

  6. I do not have a major concern about the proposal

    Tech Twitter 2% · US Tech Elite 4% · US General Population 5%

Question 7

What comes first when China is in the picture?

“If slower US frontier AI development could increase the chance that China gains ground in AI capabilities, which priority should come first?”

Tech Twitter n=1,884 · US Tech Elite n=1,526 · US General Population n=1,081

  1. Maintain a US lead even if that means moving faster

    Tech Twitter 26% · US Tech Elite 13% · US General Population 12%

  2. Balance maintaining a US lead with stronger safety requirements

    Tech Twitter 52% · US Tech Elite 55% · US General Population 53%

  3. Apply stronger safety requirements even if the US lead narrows

    Tech Twitter 15% · US Tech Elite 21% · US General Population 24%

  4. Seek international agreements before deciding

    Tech Twitter 4% · US Tech Elite 8% · US General Population 8%

  5. Neither should take priority over the other

    Tech Twitter 2% · US Tech Elite 3% · US General Population 2%

Question 8

How far should international coordination go?

“What is the furthest level of international coordination on frontier AI that you would support?”

Tech Twitter n=1,884 · US Tech Elite n=1,526 · US General Population n=1,081

  1. Broad limits on the overall pace of frontier AI development

    Tech Twitter 6% · US Tech Elite 6% · US General Population 11%

  2. Limits on how quickly AI systems can be used to improve future AI systems

    Tech Twitter 14% · US Tech Elite 18% · US General Population 17%

  3. Common testing for severe risks before frontier systems are released

    Tech Twitter 33% · US Tech Elite 35% · US General Population 32%

  4. Bans on clearly dangerous AI uses, such as helping create biological weapons

    Tech Twitter 36% · US Tech Elite 35% · US General Population 36%

  5. No international coordination on frontier AI

    Tech Twitter 11% · US Tech Elite 6% · US General Population 4%

Question 5

Should democracies adopt common safety standards?

“Do you support or oppose frontier AI companies in democratic countries adopting common safety standards with government involvement?”

Tech Twitter n=1,884 · US Tech Elite n=1,526 · US General Population n=1,081

  1. Strongly support

    Tech Twitter 28% · US Tech Elite 40% · US General Population 47%

  2. Somewhat support

    Tech Twitter 48% · US Tech Elite 42% · US General Population 36%

  3. Neither support nor oppose

    Tech Twitter 12% · US Tech Elite 9% · US General Population 10%

  4. Somewhat oppose

    Tech Twitter 10% · US Tech Elite 7% · US General Population 6%

  5. Strongly oppose

    Tech Twitter 3% · US Tech Elite 2% · US General Population 2%

What did the simulated people actually say?

Every persona's answers to all ten questions, with the reason each gave, in the response table from Radiant. Pick a society, search the text, filter by role, or focus one question to filter by its answers.

Questions

Loading responses…

Tech Twitter: responses, one row per persona.
Respondent

Question 1

When advanced AI systems may create serious societal risks, which approach to further capability development comes closest to your own view?

Question 3

Do you support or oppose requiring frontier AI companies to give independent external evaluators ongoing access to assess safety practices and report serious incidents?

Question 10

Who should have the primary authority to decide whether safety evidence is strong enough for a frontier AI company to make further capability advances?

Question 6

Do you support or oppose government-backed limits on further frontier AI capability advances when safety evidence has not kept pace?

Question 2

Overall, how do you react to Amodei's proposal to pace frontier AI development?

Question 4

How likely is it that external evaluators with employee-like access inside frontier AI companies would remain independent in their oversight?

Question 9

Which concern, if any, is most important to you when considering Amodei's proposal?

Question 7

If slower US frontier AI development could increase the chance that China gains ground in AI capabilities, which priority should come first?

Question 8

What is the furthest level of international coordination on frontier AI that you would support?

Question 5

Do you support or oppose frontier AI companies in democratic countries adopting common safety standards with government involvement?

Loading responses…

Run your own simulation in Radiant