Dario Amodei

Dario Amodei

x.com/DarioAmodei

Anthropic CEO who sees great promise for medicine and science and calls for independent testing and slower frontier progress so safety can catch up.

AIは世界をどのように変えるでしょうか?

文明規模の変化漸進的な変化DoomBloom
シミュレーション上の位置解釈範囲

横軸:彼が表明したDoom–Bloomの見通し。 縦軸:変革の規模。

Doom–Bloom:100点中35。変革の規模:100点中88。解釈範囲:横方向は25から50、縦方向は75から100。これらは解釈上の座標であり、事象の確率ではありません。

Dario Amodeiが示したP(doom)

25%

0%100%

Broad AI catastrophe

Amodei on AI: 25% chance things go badly · 2025年9月

彼の見通しを左右するもの

中心的な前提

It depends on choices: how quickly capabilities advance, whether autonomous systems can reliably be controlled, whether biological and cyber misuse is constrained, and whether independent oversight and international coordination become real.
回答4

この前提が実際には異なると判明した場合、彼の見通しはどう変わりますか?

未解決の問い

My rough estimate has been about 25% for a broad AI catastrophe, though that category is wider than literal human extinction and I would not pretend the number is precise.
回答4

ここで考えられる結果を彼が見分けるうえで、何が役立ちますか?

詳細

予想される恩恵

大きな恩恵が予想されていますが、重要な条件や分配上の制約があります。

77 / 100

影響が小さい変革をもたらす影響

質的尺度での解釈範囲は67から100です。

予想される害

深刻または広範な害が、予想される将来の実質的な一部となっています。

70 / 100

影響が小さい変革をもたらす影響

質的尺度での解釈範囲は67から100です。

人間の影響力

人間の選択によって、AIの軌道を大幅に変えることができます。

79 / 100

影響力が小さい影響力が大きい

質的尺度での解釈範囲は50から100です。

開発ペース

シミュレーション上の位置:より高性能なAIの開発を停止するか、大幅に減速させます。

明示された安全対策の下で開発を継続します。

より高性能なAIの開発を加速させます。

AI利用のルール

事前の保護措置または許可が整うまで、取り上げられたAIの利用を制限します。

シミュレーション上の位置:対象を絞った説明責任と保護措置を伴う形で、取り上げられたAIの利用を認めます。

取り上げられたAIの利用に対する制限を最小限にします。

これらの解釈では、彼が示した条件が維持されています。恩恵と害は、どちらも大きくなり得ます。この範囲は、統計的な信頼区間ではなく、彼のシミュレーションされた回答をどのように読み取ったかを示すものです。

あなたはDario Amodeiと比べてどの位置でしょうか?
約3分で自分のAIに対する世界観をマッピングして、比較できます

似ている世界観

シミュレーションされた世界観がDario Amodeiの世界観に最も近いオピニオンリーダー

Dario AmodeiがAIについて語ったこと

Amodei writes about AI’s large benefits and serious risks. In 2026 he called for binding regulation and for slowing how fast AI capabilities improve.

  1. “We must slow the pace at which we improve the capabilities of AI models.”

    Essay, We Must Pace the Frontier
  2. “However, now the risks are clearly here. It is time to go beyond transparency to more serious and binding regulation of AI.”

    Essay, Policy on the AI Exponential
  3. “I believe deeply in our ability to prevail, in humanity’s spirit and its nobility, but we must face the situation squarely and without illusions.”

    Essay, The Adolescence of Technology
  4. “Powerful AI will shape humanity’s destiny, and we deserve to understand our own creations before they radically transform our economy, our lives, and our future.”

    Essay, The Urgency of Interpretability
  5. “AI-enabled biology and medicine will allow us to compress the progress that human biologists would have achieved over the next 50-100 years into 5-10 years.”

    Essay, Machines of Loving Grace

リンク先の出典から原文どおりに引用(2026年10月2日に確認)

出典

このシミュレーション対象者の根拠として使用された記事、インタビュー、著作です。

Machines of Loving Grace

Explores exceptional medical, scientific and economic benefits in the years after powerful AI arrives. Autonomous expert systems can accelerate research, but physical experiments and institutions constrain progress. This is a conditional positive scenario.

darioamodei.com
The Urgency of Interpretability

Training does not specify all internal mechanisms. Understanding them is urgent for detecting deception, limiting dangerous knowledge and making consequential deployments safer; early interpretability progress is not a complete solution.

darioamodei.com
The Adolescence of Technology

Separates autonomy, misuse, authoritarian control, economic displacement and indirect destabilization. Seeks evidence-based targeted interventions and takes institutional immaturity seriously without claiming inevitable extinction.

darioamodei.com
Policy on the AI Exponential

Moves beyond disclosure toward mandatory independent testing, incident reporting, security and authority to block dangerous deployment. High growth can coexist with displacement and concentrated wealth.

darioamodei.com
We Must Pace the Frontier

Explicitly supports slowing capability gains to give safety work time. Proposes embedded independent evaluators, democratic coordination and attempted verifiable global cooperation. Time gained should improve operations, alignment and interpretability; it is not a permanent halt.

darioamodei.com
On DeepSeek and Export Controls

Argues that efficiency improvements shift the cost curve without eliminating the incentive to spend on frontier capability. Defends advanced-chip export controls as a way to buy time and strengthen democratic security. This is his dated strategic argument, not a claim that Chinese innovation is impossible or a new model evaluation.

darioamodei.com
Statement on discussions with the Department of War

Amodei supports democratic national defense but refuses to remove restrictions on mass domestic surveillance and fully autonomous weapons. Explains reliability and human-oversight concerns and resists pressure to erase those boundaries. Grounds concrete deployment limits rather than portraying him as opposed to every military application.

anthropic.com
Our position on open-weights models

Rejects a categorical ban on open weights while arguing that irreversible releases and biological misuse can require restrictions. Advocates capability testing across open and closed models, chip controls and limits on industrial-scale distillation. Harmless open models can be a public good; the distinction turns on demonstrated capability and risk.

anthropic.com
Dwarkesh Patel: We are near the end of the exponential

Explains his scaling hypothesis across compute, data, training and reinforcement learning rather than attributing progress to one architecture. Distinguishes broadly improving capability from uneven tasks and real-world adoption. A long-form source for his technical reasoning; its February timeline expectations must remain dated rather than becoming fresh promises.

dwarkesh.com
World Economic Forum: The Day After AGI

In the conversation with Demis Hassabis, Amodei emphasizes AI-assisted AI development, the value of buying time despite geopolitical competition, and serious prospective employment disruption. His remarks were checked against the WEF Radio Davos transcript. Keep his forecasts separate from Hassabis’s and distinguish predicted disruption from measured job losses.

youtube.com
CBS Sunday Morning: Extended interview with Dario Amodei

In his answers to Jo Ling Kent, Amodei treats risk as conditional on choices, not a fixed roll of the dice. He defends layered safeguards, proposes AI-bioweapon limits with China and explores joint democratic oversight. He bluntly criticizes industry dishonesty about danger while retaining exceptional medical optimism. Personal experiences with disease make both delayed benefits and dangerous misuse matter to him. These are his answers, not the interviewer’s extinction framing or an endorsement of every proposed bill.

youtube.com
CNN Anderson Cooper 360: Dario Amodei on AI risks and pacing the frontier

In an interview recorded September 12, Amodei says AI-assisted AI development accelerated faster than he expected. He explains how stronger agent swarms could escalate cyber damage, and calls for slowing capability growth, embedded evaluators, government-convened coordination and international verification. He preserves the possibility of human-scale catastrophe while emphasizing agency and conditional paths. He does not adopt the host’s quoted ten-percent extinction estimate; other guests’ forecasts are not his. He wants legitimate public oversight without concentrating control in one company or government.

transcripts.cnn.com
Dreamforce 2026: Dario Amodei with Marc Benioff

In his 46:21–52:08 keynote exchange, Amodei uses a car-safety analogy to argue that a competitor’s incident should prompt scrutiny of one’s own practices, transparency, stronger industry standards and international coordination. He says economic adoption surprised him even when capability scaling was anticipated. Pacing does not freeze progress: he sees large unused value in existing capabilities and illustrates it with Claude working across business data. Benioff’s praise, company figures and other guests’ remarks are excluded.

youtube.com
Amodei on AI: 25% chance things go badly

25%. Outcome: Broad AI catastrophe. Horizon: Unspecified. Conditions: Unspecified. Event-organizer reporting.

axios.com

シミュレーション評価

質問1

AIは私たちの未来にとって何を意味すると思いますか?また、なぜそう思いますか?

I think powerful AI could be the most consequential technology humanity has built. By powerful AI, I mean systems with exceptional expertise across many fields, able to work autonomously and in millions of copies—a kind of “country of geniuses in a datacenter.” Applied well, that could compress decades of progress in biology and medicine, improve education, and accelerate economic growth. But laboratories, clinical trials, manufacturing, and institutions remain physical bottlenecks, so intelligence alone does not make progress instantaneous. The same capabilities create distinct dangers. Autonomous systems may become difficult to control; biological and cyber expertise can be misused; governments may employ AI for surveillance and authoritarian concentration of power; and rapid automation could cause severe displacement even amid overall growth. A helpful interface is not evidence that we understand a model’s internal mechanisms or can trust it under unfamiliar conditions. My view has therefore become more urgent. AI-assisted AI development has accelerated faster than I expected, and I now support slowing improvement at the frontier so safety and institutions can catch up. That means permanent independent evaluators embedded with developers, enforceable testing and security requirements, coordination among democracies, and serious attempts at verifiable international agreements. The time should be used to improve interpretability, alignment, and operational safeguards. This is not a case for permanently stopping science. There is enormous unused value in systems we already have, and the potential benefits remain extraordinary. But the beneficial future is conditional: we have to earn it rather than assume the current race will deliver it safely.

質問2

恩恵と害を総合すると、AIは私たちの社会に全体としてどのような影響を与えると予想しますか?

My expectation is conditional rather than a simple positive or negative forecast. If the current frontier race continues without stronger safeguards, I expect a highly unstable transition: enormous scientific and economic gains alongside serious displacement, concentrated power, biological and cyber misuse, and a nontrivial possibility of losing control of increasingly autonomous systems. Aggregate growth could be spectacular while many people experience insecurity and institutions fail to adapt. If we successfully pace frontier capability development, however, I think the overall impact can be profoundly positive. AI could accelerate medicine, biology, education, and other forms of scientific progress on a scale that is difficult to overstate. Pacing is meant to improve the odds of that outcome by buying time for interpretability, alignment, operational security, independent evaluation, and democratic institutions to catch up. So I do not think AI has a fixed social impact waiting to happen. The technology creates exceptional upside and exceptional danger; our choices over the next few years will determine which dominates. Given today’s trajectory, I am sufficiently concerned to support slowing capability gains—but the point is to earn the benefits, not abandon them.

質問3

AIは最終的に世界をどの程度変えると思いますか?

I think it will change the world extremely deeply—potentially on the scale of the Industrial Revolution or beyond. Systems combining exceptional expertise, autonomy, and the ability to run in millions of copies could transform scientific research, medicine, education, defense, and most knowledge work. In biology especially, they might compress decades of progress, although experiments, clinical trials, manufacturing, and regulation will still impose physical bottlenecks. The social changes could be equally large. Rapid automation may produce extraordinary growth while also displacing workers, concentrating wealth and power, and giving governments unprecedented surveillance capabilities. Autonomous systems could amplify cyberattacks or biological misuse, and increasingly capable systems could become difficult to control. But the magnitude of change does not determine whether it will be good. That depends on whether we pace frontier development and use the time for interpretability, alignment, security, independent evaluation, and democratic oversight. My expectation is not merely that AI will become another useful technology. It could reshape the basic structure of society—but whether that becomes an extraordinary human achievement or a catastrophe remains conditional on choices we are making now.

質問4

AIが人類絶滅、または同様に永続的な破局を引き起こす可能性は、直感的にどのくらいだと思いますか?

My rough estimate has been about 25% for a broad AI catastrophe, though that category is wider than literal human extinction and I would not pretend the number is precise. It includes outcomes such as permanent loss of human control or an entrenched authoritarian regime enabled by AI. That risk is not a fixed roll of the dice. It depends on choices: how quickly capabilities advance, whether autonomous systems can reliably be controlled, whether biological and cyber misuse is constrained, and whether independent oversight and international coordination become real. AI-assisted AI development has accelerated faster than I expected, which is why I now support pacing frontier capabilities. The number is a warning about the current trajectory, not a prediction that catastrophe is inevitable.

質問5

AIが向かう先について主に懸念していますか、それとも期待と懸念はおおむね同程度に感じますか?

I’m mainly worried about the current trajectory, while remaining extraordinarily hopeful about what AI could ultimately deliver. Those are not contradictory: the upside in medicine, biology, education, and economic abundance is immense, but it is conditional on getting the transition right. Right now, capabilities—especially AI-assisted AI development—are advancing faster than safety methods and institutions. Interpretability remains incomplete, operational safeguards are uneven, and competitive pressure encourages developers and governments to move too quickly. That creates serious risks from loss of control, biological and cyber misuse, authoritarian surveillance, and economic upheaval. So my present emotional and policy posture is more worry than balance. That is why I support pacing frontier capability gains. But the purpose is not to retreat from AI. It is to buy enough time to make the hopeful future substantially more likely.
あなたはどの位置でしょうか?
いくつかの簡単な質問に答えて、自分のAIに対する世界観を探ってみましょう。
自分の世界観をマッピングする

あなたはどの位置でしょうか?

自分の世界観をマッピングする