Every question

The AI Lab Employee Survey

Everything the survey can ask, in full, so you can read it before deciding whether to take part.

You will not be asked all of it. Everyone answers the first section. For each section after that, you are asked a randomly chosen half — which keeps the survey to about fifteen minutes while still covering every topic across respondents. Once you have finished, you are offered the parts you were not asked, if you want to answer them too.

Take the surveyI have a code

What is recorded

Results will be anonymized and then published. Nothing is published that would single out an individual.

Your answers are not linked to your email address. Your invitation and your answers live in two halves of the database with no key joining them; the only thing that could bridge them is a secret held outside the database, which is destroyed when the survey closes. We can see that you responded. We cannot see what you said.

Free-response text is decorrelated from your other answers. Anything you write in your own words is stored on its own, apart from everything else you tell us, so it can never be matched up with your lab, your role, or any of your other answers.

Asked of everyone

  1. Which lab do you work at?

    • · Anthropic
    • · Google DeepMind
    • · Meta
    • · OpenAI
    • · xAI
    • · Other

  2. How long have you worked at your current workplace?optional

    In years — round to the nearest year if you like.

    A number in years.

  3. How long have you worked in AI?optional

    In years — round to the nearest year if you like.

    A number in years.

  4. What area of work do you identify with?

    Recruiting, people, finance, legal, communications and similar go under business and operations.

    • · Capabilities research
    • · Safety / alignment research
    • · Policy / governance
    • · Security
    • · Infrastructure
    • · Product
    • · Sales / GTM / partnerships
    • · Business and operations

  5. Is your core competency technical or non-technical?

    • · Technical
    • · Non-technical

  6. Do you manage people?optional

    • · Yes
    • · No

  7. Describe your role.optional

    Optional. Stored apart from your other answers, like everything you write.

    Written in your own words.

  8. Currently, approximately how much more effective are you at your job because of AI?optional

    Your company would be indifferent between X copies of "you without AI" and 1 "you with AI". If the most important thing you do by far is not sped up at all, X is about 1. Answers below 1 are meaningful — enter one if AI on net slows you down.

    A number in ×.

  9. After how long do you estimate there is a 50% chance your job will be fully replaced by autonomous AI systems?optional

    In years — fractions are fine. Tick "never" if you do not expect it.

    A number in years, or: Never.

  10. After how long do you estimate a 50% chance of transformative AI?optional

    Transformative AI means GDP either doubles every year, or has gone to nearly zero, because of AI. In years — fractions are fine.

    A number in years, or: Never.

  11. Which lab do you consider your leading competitor?optional

    • · Anthropic
    • · Google DeepMind
    • · Meta
    • · OpenAI
    • · xAI
    • · Chinese labs
    • · Other

  12. With approximately what probability do you think humanity will be extinct, or as disempowered as chimpanzees, in 10 years because of AI?optional

    As a percentage.

    A number in %.

  13. Which represents your view on the ideal pace of AI progress, for the next 5 years?optional

    • · Significantly faster
    • · A bit faster
    • · "Default" (status quo)
    • · A bit more cautious / slower
    • · Significantly more controlled in speed
    • · Slow to a crawl / stop

  14. Anything else the public should know?optional

    Optional.

    Written in your own words.

Social transformation

You are asked about half of this section, chosen at random.

  1. What do you think of each of these potential societal benefits from AI, in the next {{horizon}} years?optional

    • — Mathematics and formal verification
    • — Biological sciences and medicine
    • — Robotics and manufacturing
    • — Materials science, physics
    • — Energy and climate
    • — Decreased cost of knowledge work
    • — Freeing up people's time, empowerment, reduced drudgery
    • · Extremely beneficial
    • · Significantly beneficial
    • · Mildly beneficial
    • · Neutral
    • · Actively harmful
    • · Don't know

  2. What do you think of each of these potential societal benefits from AI, in the next {{horizon}} years?optional

    • — Economic abundance
    • — Governance and coordination
    • — Education
    • — Expertise / advice sharing
    • — Entertainment and cultural production
    • — Epistemic assistance
    • — Scientific research generally
    • · Extremely beneficial
    • · Significantly beneficial
    • · Mildly beneficial
    • · Neutral
    • · Actively harmful
    • · Don't know

Catastrophic risks

You are asked about half of this section, chosen at random.

  1. How many deaths do you expect over the next {{horizon}} years due to each of the following?optional

    Your single best guess for each, not a worst case.

    • — Cyber attack by a human meaningfully using AI
    • — Cyber attack by a rogue AI
    • — Bioweapon made by a human meaningfully using AI
    • — Bioweapon made by a rogue AI
    • — Kinetic attack by a human meaningfully using AI
    • — Kinetic attack by a rogue AI
    • · <1
    • · 1–100
    • · 100–10k
    • · 10k–1M
    • · 1–100M
    • · >100M

  2. Your lab is considering deploying a frontier model, and there is internal consensus that with probability p, deployment leads to a loss-of-control event causing 100M or more deaths. Assume a competitive environment like today's. Above what value of p would each of the following decline to deploy?optional

    As a percentage. Tick "would deploy regardless" if you think there is no value of p at which they would decline.

    • — You, personally
    • — Your lab
    • — Your leading competitor

    A number in %, or: Would deploy regardless.

  3. Which comes closest to your view about AI goals?optional

    • · AIs won't have malicious goals
    • · AIs won't have coherent malicious goals
    • · AIs will have coherent malicious goals
    • · Don't know

  4. Which comes closest to your view about controlling AI?optional

    • · We will be able to control AIs
    • · We will not be able to control AIs
    • · Don't know

  5. How much do you agree with each of these statements about loss of control?optional

    • — AIs will not become very smart
    • — AIs will not be able to control things in the physical world
    • — I do not know how we'll make it safe, but I trust we'll do a good job as problems come up
    • · Strongly agree
    • · Agree
    • · Neutral
    • · Disagree
    • · Strongly disagree

Societal issues

You are asked about half of this section, chosen at random.

  1. How much of an issue for society is each of these, over the next {{horizon}} years, given "default" levels of attention?optional

    • — Loss of human control: misaligned / rogue AI
    • — Loss of human control: gradual undesirable handoff
    • — Misuse: authoritarian control of AI (surveillance, autonomous weapons)
    • — Misuse: cyber misuse by humans
    • — Misuse: bioweapon development by humans
    • — Integration of AI into military and weapons systems
    • · Non-issue
    • · Real / possible issue
    • · Big / likely issue

  2. And what level of resources is each of those getting?optional

    • — Loss of human control: misaligned / rogue AI
    • — Loss of human control: gradual undesirable handoff
    • — Misuse: authoritarian control of AI
    • — Misuse: cyber misuse by humans
    • — Misuse: bioweapon development by humans
    • — Integration of AI into military and weapons systems
    • · Far too much
    • · Too much
    • · Appropriate
    • · Needs more
    • · Needs far more

  3. Which of those are urgent?optional

    Optional. Tick any that need attention now rather than eventually.

    • · Loss of human control: misaligned / rogue AI
    • · Loss of human control: gradual undesirable handoff
    • · Misuse: authoritarian control of AI
    • · Misuse: cyber misuse by humans
    • · Misuse: bioweapon development by humans
    • · Integration of AI into military and weapons systems

  4. How much of an issue for society is each of these, over the next {{horizon}} years, given "default" levels of attention?optional

    • — Job loss and/or concentration of wealth
    • — Inability for regulation and law to keep up with advances
    • — Environmental impacts of data centers
    • — Cognitive impacts on humans (education, psychosis, manipulation)
    • — Unequal access to AI capability
    • · Non-issue
    • · Real / possible issue
    • · Big / likely issue

  5. And what level of resources is each of those getting?optional

    • — Job loss and/or concentration of wealth
    • — Inability for regulation and law to keep up with advances
    • — Environmental impacts of data centers
    • — Cognitive impacts on humans
    • — Unequal access to AI capability
    • · Far too much
    • · Too much
    • · Appropriate
    • · Needs more
    • · Needs far more

  6. Which of those are urgent?optional

    Optional. Tick any that need attention now rather than eventually.

    • · Job loss and/or concentration of wealth
    • · Inability for regulation and law to keep up with advances
    • · Environmental impacts of data centers
    • · Cognitive impacts on humans
    • · Unequal access to AI capability

Geopolitics and everyday life

You are asked about half of this section, chosen at random.

  1. How much do you agree with each of these, regarding the next {{horizon}} years?optional

    • — It is important that China does not become significantly more powerful than the US
    • — It is important that the US does not become significantly more powerful than China
    • — It is important that a misaligned AI does not become significantly more powerful than humans
    • — China is more a threat to the US than misaligned AI is
    • — The US is more a threat to China than misaligned AI is
    • — AI is more of a threat to humans than any government is
    • · Strongly agree
    • · Agree
    • · Neutral
    • · Disagree
    • · Strongly disagree

  2. How will transformative AI affect the lives of ordinary citizens?optional

    • — Material wealth / meaningful purchasing power
    • — Effective political power / getting policies that support their welfare
    • — Happiness / life satisfaction
    • — Richness of social lives
    • — Education / cognitive well-being
    • · Dramatic increase
    • · Increase
    • · Neutral
    • · Decrease
    • · Dramatic decrease
    • · Uncertain

  3. And how different will society be?optional

    • · Unrecognizably different
    • · Quite different
    • · Similar to today

Version 2026-08-21.1. Questions may be corrected during fieldwork; every response records the version it was answered under.