ClaimsScott Alexander

“In our scenario, that starts with cyber warfare, the government sees that these AIs are now as capable as the best human hackers, but can be deployed at humongous scale.”

Scott Alexander, co-author of AI 2027 3 April 2025 AI 2027: month-by-month model of intelligence explosion — Scott Alexander & Daniel Kokotajlo (Dwarkesh Podcast) archived copy

Prediction AI policy and governance

In the AI 2027 scenario, the government sees via cyber warfare that AIs are as capable as the best human hackers but deployable at huge scale.

Mostly true True in substance, wrong in a detail.

Prediction: Statements about events outside the person's control, checked once they fell due: did they come true?

What would count as true or false

This is a claim about what the AI 2027 scenario says, not about real-world events. True: the scenario text describes the government's recognition of AI's strategic importance starting with, or driven by, cyber capabilities, with AIs matching top human hackers and able to run at huge scale (many copies at once). Mostly true: cyber capability matching top hackers and deployable at scale is shown as a main driver of the government's recognition, but not clearly as the first. Mixed: the scenario mentions AI cyber capabilities, but the government's recognition is driven mainly by other factors, or one of the two elements (human-expert level, mass scale) is missing. Mostly false: cyber capabilities appear only in passing and play no part in the government's recognition. False: the scenario contains no such development.

Set before the research began.

Evidence (4)

  1. E1 AI 2027 AI 2027 authors (ai-2027.com) · 3 April 2025 Own record

    “Officials are most interested in its cyberwarfare capabilities: Agent-2 is “only” a little worse than the best human hackers, but thousands of copies can be run in parallel, searching for and exploiting weaknesses faster than defenders can respond.”

  2. E2 AI 2027 Forecast Explained: Exploring the Path from AGI to Superintelligence Zenn (third-party article by taku_sid) · 12 April 2025

    “The National Security Council (NSC), the Department of Defense (DOD), and the U.S. AI Safety Institute (AISI) show interest. Particular attention is focused on its cyber warfare capabilities. Agent-2 is only slightly inferior to the best human hackers, but it can run thousands of copies in parallel, searching for and exploiting vulnerabilities faster than a human defender can respond.”

  3. E3 AI 2027 AI 2027 authors (ai-2027.com) · 3 April 2025 Own record

    “The Department of Defense considers this a critical advantage in cyberwarfare, and AI moves from #5 on the administration’s priority list to #2.”

  4. E4 My Takeaways From AI 2027 Astral Codex Ten (Scott Alexander) · 8 April 2025 Own record

    “AI will scare people with hacking before it scares people with bioterrorism or whatever.”

Reasoning

The scenario text says officials are most interested in Agent-2's cyberwarfare capabilities: it is "only a little worse than the best human hackers" and thousands of copies can run in parallel (E1). This leads the Department of Defense to treat it as a critical advantage, and AI moves from #5 to #2 on the administration's priority list (E3). An independent summary confirms this (E2). The scenario shows cyber capability at scale as a main driver of the government's interest [E1][E3], but it puts the AI slightly below the best human hackers rather than equal to them, and does not explicitly call this the first step.

Cite this claim

Scott Alexander, “In the AI 2027 scenario, the government sees via cyber warfare that AIs are as capable as the best human hackers but deployable at huge scale.” (said 3 April 2025). Verdict: mostly true. Reliability API, claim c_d247da1f8cba, method 0.1.0+8bea457b, dataset release v1.1.2. https://epistacklabs.com/claims/c_d247da1f8cba

This claim is in dataset release v1.1.2, with the same verdict.

Reply to this claim

For Scott Alexander, or someone speaking for them: a correction or counter-evidence. Replies are read before anything is shown; an approved reply shows your name, role, date, statement and link. Your email address is never shown and is deleted 30 days after the decision (privacy).

More prediction claims by Scott Alexander

  1. Mostly true Prediction 3 April 2025
    “We expect that as the AI labs become more capable, they tell the government about this because they want government contracts, they want government support.”

    In the AI 2027 scenario, as AI labs become more capable they tell the government, seeking government contracts and support.

  2. Resolved true Prediction 3 April 2025
    “So they become extremely interested and they discuss nationalizing the AI companies.”

    In the AI 2027 scenario, the government becomes extremely interested in AI and discusses nationalizing the AI companies.

  3. Mostly true Prediction 3 April 2025
    “Our scenario focuses on coding in particular because we think coding is what starts the intelligence explosion.”

    The authors of the scenario think coding is what will start the intelligence explosion.

  4. Mostly true Prediction 20 February 2023
    “I expect scaling to continue to produce gains, but this is less obvious than it’s been the past five years.”

    Scaling AI models will continue to produce gains, though less obviously than in the past five years.

  5. Resolved true Prediction 20 February 2023
    “Training GPT-4 will cost $100M, which is a lot.”

    Training GPT-4 will cost $100 million.

All of them

Verdicts are made automatically; how verdicts are made.