Who he is
Affiliation: Co-founder, Ethereum; board member, Ethereum Foundation — role check run 2026-09-21: the EF restructured in early 2025 (Aya Miyaguchi to President; Hsiao-Wei Wang and Tomasz Stańczak as co-executive directors, announced 2025-01/03), and through 2026 shed staff (~20% cut after a 40% budget reduction) and senior names (Stark, Beiko, Monnot; Wang stepped down mid-2026, Bastian Aue interim ED). Buterin, Miyaguchi, and Storchenegger remain the board (EF leadership tracker, 2026). He deliberately reduced his operational influence — independent researcher / essayist (vitalik.eth.limo) is now the accurate primary description.
One-line position: AI x-risk is real, but a moratorium is both infeasible and dangerous; the right path is d/acc — differentially accelerate defensive, decentralized, democratic technologies so society can absorb powerful AI without ceding control to a few labs or states. New since v1: he now says it plainly as a speed preference — "ride through the narrow corridor at 60 kph and not 200 kph" (X thread, 2026-07-20).
Discipline & technical bet
Protocol designer and mechanism-design essayist — the only person on the bench who has shipped civilizational-scale infrastructure and also engages x-risk arguments on their merits. Current technical bets: (1) Ethereum as the coordination substrate for the AI economy — private model interactions, anonymous payments for AI services, cryptographic verification of AI behavior, agent-to-agent economic coordination via deposits/reputation/dispute-resolution (X post, 2026-02-10); (2) AI flips cybersecurity to defense-dominance via formal verification — if AI can prove hard theorems, it can prove software security properties end-to-end; the hard part is specifying "secure" (Decrypt, 2026-09-17); (3) protocol-level: the 2025–26 scaling path ran through Pectra (2025-05) and Fusaka/PeerDAS (2025-12-03), with the ZK-centric "lean Ethereum" direction next.
Key claims (Says)
- Empirical: catastrophic / existential AI risk is real — takes Bengio, Russell, and (partially) Yudkowsky seriously. Held — restated 2026: the AGI race "mirrors the unchecked speed and scale Ethereum was designed to oppose" (X, 2026-02-10); supports deliberately slowing progress (X, 2026-07-20).
- Normative: a global pause is politically infeasible and, attempted asymmetrically, concentrates capability in the least-cautious actors. Held — unchanged; note the register shift toward explicit slow-down preference (60-not-200) while still rejecting moratorium mechanics.
- Normative: the frame is differential acceleration — defensive tech (cyber defense, info verification, bio defense, decentralized infrastructure) faster than offensive / centralizing tech. Held — "choosing a positive direction rather than embracing undifferentiated acceleration of the arrow" (X, 2026-02-10).
- Empirical / structural: decentralization is a safety property, not an ideological preference ("My techno-optimism," 2023-11). Held — now given an AI-specific mechanism list (verification, private inference, distributed coordination; X, 2026-02-10).
- Empirical: open-source AI is, on net, safer than closed frontier labs over long horizons. Held — "favors open-weights models to prevent centralized dominance" (X, 2026-07-20); proliferation counter-argument still unresolved (weak spot retained).
- Empirical / mechanism: prediction markets and "info-finance" are under-used civilizational tools ("From prediction markets to info finance," 2024-11). Held on adoption — the institutions arrived: ICE invested $2B in Polymarket (ICE press release, 2025-10) and added ~$600M in 2026; Kalshi valued ~$22B (2026). The deeper falsifier — do these markets shift institutional decisions — is still live.
- Normative: democracy + markets + cryptography together are the substrate that keeps humans in the loop as AI scales. Held — extended with the "augmented governance" thread (2026).
- Empirical / contested: more sympathetic to alignment concerns than the crypto/open-source camp (Andreessen, Mostaque, LeCun) while rejecting their x-risk dismissal. Held — the 2026-07-20 thread concedes real uncertainty against his own optimism: "I am simply not certain that there actually is a significant range of human capabilities that will stubbornly resist LLMs."
- NEW (2026-07-20): Definitional: AGI = "AI powerful enough that, if uploaded into robot bodies and all humans disappeared, it could independently sustain and advance civilization" — a falsifiable-in-principle threshold, sharper than most lab definitions; frames three machine waves (industrial, computation, LLMs) and bets on human-machine symbiosis over substitution.
- NEW (2026-09-17): Empirical: AI-powered hacking does not doom security — "cybersecurity is naturally defense-favoring once people get their shit together"; AI-scale formal verification is the mechanism (Decrypt, 2026-09-17). This is d/acc's most concrete testable instance to date.
Notable predictions — with falsifiable checks
- (2013-11) Ethereum whitepaper: a smart-contract platform will exist alongside Bitcoin. Held — Ethereum launched 2015-07-30 and held the #1 smart-contract position through the 2025–26 upgrade cycle (Pectra 2025-05; Fusaka 2025-12-03). Check: still #1 by TVL/activity vs Solana at next review.
- (2016+) Proof-of-stake achievable and superior. Held — The Merge, 2022-09-15; no consensus failure through 2026.
- (2022-09) ZK-rollups the dominant scaling path. Held (directionally) — L2 ecosystem matured; Fusaka/PeerDAS (2025-12-03) shipped the data-availability layer the rollup roadmap needs; ZK-centric "lean Ethereum" is the stated next phase (Glamsterdam). Check: ZK-rollup share of L2 activity, end-2027.
- (2023-07) Worldcoin biometric-orb critique on privacy/coercion grounds. Held (partially vindicated) — regulatory pushback in Kenya, Germany, Spain, Portugal, UK (2023–24). Check: 2026–27 jurisdictional status not re-verified this cycle; catalog next review.
- (2023-11-27) d/acc becomes a recognized third position between e/acc and pause. Drifted — the term entered AI-discourse vocabulary and he keeps restating it (revisited in the 2026-02-10 post; "d/acc: one year later," 2025-01), but no political coalition formed, and the 2026 axis of debate became US-vs-China / preemption, not e/acc-vs-d/acc — the third position exists on paper, not in power. Check: any institution, caucus, or fund explicitly organized around d/acc by end-2027; if none, regrade Wrong.
- (2024-11) Info-finance / prediction markets as governance infrastructure. Held on volume and institutionalization — ICE $2B (2025-10) + ~$600M follow-on; Kalshi ~$22B. Falsifier still live: a documented case of a prediction market changing an institutional decision by end-2027.
- (2023–25) Decentralized AI compute / inference becomes a viable alternative to hyperscaler dominance. Drifted — the 2026 reality is the opposite trade: a trillion-dollar centralized infra buildout (see Andreessen Machine Age entry) with no decentralized-inference breakout at competitive economics. Check: any decentralized inference network serving >1% of frontier-model-class traffic by end-2027; if not, regrade Wrong.
Revealed behavior (Does)
- Deliberately shrank his own institutional power: initiated the EF restructure (2025-01), moved to board-only oversight, and let the foundation cut budget ~40% and staff ~20% — the only bench member who responded to "concentration is dangerous" by de-concentrating himself.
- Time allocation went to research and writing: the 2025–26 output is essays and X threads on AI trajectory, governance, and security — not fund-raising, not a startup, not a policy shop.
- Protocol milestones kept shipping on his roadmap (Pectra 2025-05, Fusaka 2025-12-03) despite the EF's institutional churn — the technical program survived the organizational diet.
- Still no equity positions in AI labs or funds found; wealth remains concentrated in ETH — the COI profile is unchanged in kind since v1.
Feels
Fear of lock-in above all — permanent concentration of power reads to him as worse than slow progress or even moderate catastrophe risk. Wants to be true: that symbiosis beats substitution, and that a civilization can steer "the arrow" without a sovereign hand on it.
Hears
Ethereum core devs and the cryptography research world, mechanism-design and EA-adjacent forums, rationalist safety writers (reads Yudkowsky and the labs' safety output seriously), effective-accelerationist critics he answers by name; notably light on Washington and on frontier-lab commercial circles.
Sees
A live, decade-long experiment in decentralized governance at billion-dollar stakes — forks, MEV, governance attacks, treasury politics — giving him mechanism-level intuitions about how distributed systems actually fail that no academic or lab CEO on the bench possesses.
Incentive map
Re-derived this cycle. Holds substantial ETH; "decentralization is a safety property," "Ethereum as AI coordination substrate," and "info-finance" all appreciate his own asset — structural similarity to Andreessen's portfolio-talk, from the opposite ideological pole. The ICE/Polymarket institutional wave also validates (and enriches) the info-finance ecosystem he championed. Offsetting evidence: he cut his own institutional power at the EF, and his x-risk sympathy runs against crypto-tribal interest. Cannot easily say: "Ethereum is the wrong substrate for AI coordination" or "decentralized inference is economically dead." Rule-25 discount: moderate — high on Ethereum-adjacent claims, low on his AI-risk positions, which cut against his commercial tribe.
Theories aligned with
- d/acc (defensive / decentralized / differential / democratic accelerationism) — his own framework ("My techno-optimism," 2023-11-27; "d/acc: one year later," 2025-01) [TBD: needs own theory file at ../theories/d-acc.md]
- Techno-optimism — the conditional, defensively-gated variant; explicitly distances from Andreessen's unconditional version
- AI alignment / x-risk — accepts the empirical concern, splits on the political response
- Cypherpunk / crypto-anarchism lineage (Chaum, May, Szabo) — operational, not ideological
What he's reacting against
- Marc Andreessen's "Techno-Optimist Manifesto" frame (2023-10) that treats all acceleration as good and dismisses x-risk as moral panic
- Yudkowsky-style pause/moratorium framing — argues it concentrates power in whoever ignores the moratorium
- Frontier-lab self-governance as the load-bearing safety mechanism — structurally fragile
- Crypto-libertarian framing that treats decentralization as an end rather than a defensive property
- "AI safety" discourse that ignores the political economy of who ends up holding the levers
- NEW: AI-hacking doomerism in security discourse — the 2026 target is fatalism about offense-dominance (Decrypt, 2026-09-17)
Where he overlaps / splits (with Rich)
- Splits with Andreessen — "My techno-optimism" is the explicit response; agrees on direction, rejects x-risk dismissal and unconditional acceleration. The 2026 contrast sharpened: Andreessen raised $9.6B for centralized AI infra while Buterin argued for slowing and decentralizing the same buildout.
- Splits with Yudkowsky — agrees x-risk is real; rejects moratorium as infeasible and itself a concentration mechanism. Cleavage is over response, not diagnosis.
- Overlaps with Bengio, Russell, Hinton, Toner on x-risk and governance need; splits on lever — they want institutions and accountability; he wants differentially accelerated defensive technology + open infrastructure.
- Overlaps with Mostaque on open-source / decentralization as safety; splits by being more programmatic and less crypto-utopian.
- Overlaps with LeCun on open weights; splits sharply on x-risk — LeCun dismisses, Buterin takes seriously.
- Overlaps with Amodei, Hassabis, Suleyman on safety mattering; splits on lab concentration being itself part of the safety problem.
- Splits with Acemoglu, Varoufakis — shares power-concentration concern, reaches for cryptographic / market / protocol fixes rather than state institutions or democratized cloud capital.
- Distinct from Thiel — both anti-stagnation; Thiel wants determinate monopoly-building, Buterin wants permissionless substrate.
Track record
- Ethereum (2013–present): second-largest crypto network; survived boom/bust cycles, the DAO fork, a consensus transition, and (2025–26) its own foundation's downsizing. Direct execution record, not commentary.
- Technical writing: cross-domain essays (cryptography, governance, mechanism design, AI) that age comparatively well and are cited across tribes.
- Worldcoin critique (2023): privacy-and-coercion concerns broadly tracked by subsequent regulatory action.
- d/acc (2023-11): entered the vocabulary, cited by safety researchers; influence on actual policy or lab behavior remains weak — his one framework-level prediction is the one drifting.
- Info-finance (2024-11): called the institutionalization early — ICE/Polymarket and Kalshi's 2025–26 wave arrived within 18 months.
- Pattern: strong on identifying structural / mechanism problems early; weaker on operational specifics outside crypto (d/acc taxonomy is direction, not implementation).
Empirical vs normative
- Empirical: x-risk non-trivial; lab concentration fragile; open-source safer long-run (contested); prediction markets surface information; defense-favoring security via formal verification; decentralization raises capture costs
- Normative: differential acceleration is the right frame; defensive tech should be subsidized; democratic + cryptographic substrate; permanent lock-in is the worst outcome — worse than slow progress
- Conflict note (CONVENTIONS rule 25): substantial ETH exposure aligned with the substrate claims; discount those, but note his AI-safety positions cut against his commercial tribe — the COI is asymmetric across claim types
Weak spots / open questions
- d/acc is a taxonomy, not an implementation — the category-to-deployable-system gap remains after three years; the formal-verification security bet (2026-09) is the first concrete instance worth tracking as a test case
- "Decentralization is a safety property" is under-argued for the AI-specific case — misalignment/misuse failure modes aren't obviously reduced by distributing weights; possibly increased via proliferation. Acknowledged but unresolved.
- COI on decentralization claims — Ethereum founder arguing decentralized infrastructure is safer parallels a16z portfolio bias from the other pole
- Info-finance → AI-control leap — "Polymarket aggregates information" to "prediction markets help align AI" is large and under-specified; ICE's $2.6B validates the market, not the governance thesis
- Open-source-safer bet genuinely contested — bio-proliferation especially; rests on "concentration is worse" rather than rebutting proliferation directly
- Light on frontier-lab technical safety specifics — RLHF, RSPs, interpretability, evals; d/acc complements lab-internal safety work, doesn't substitute for it
- Falsifiability: most d/acc claims are directional; the checks attached above (coalition test, decentralized-inference share, verification test case) are this file's attempt to make him gradeable — hold him to them
Rich's take
- (your synthesis here)
Delta log
2026-09-21 — v1→v2 migration + validation (weekly batch)
- Grades: 13 Held / 2 Drifted / 0 Wrong.
- Held: x-risk-real (restated X post via CoinDesk, 2026-02-10); pause-infeasible; d/acc frame; decentralization-as-safety; open-weights preference (X thread, 2026-07-20); info-finance (ICE $2B into Polymarket, ICE, 2025-10, plus ~$600M follow-on and Kalshi ~$22B, 2026); democracy-substrate; alignment-sympathy; Ethereum-dominance; PoS; ZK-scaling (Fusaka 2025-12-03); Worldcoin critique (partial); defense-favoring-security thesis stated as new claim (Decrypt, 2026-09-17).
- Drifted: d/acc-as-third-position (vocabulary yes, coalition no — the 2026 axis became US-vs-China, not e/acc-vs-d/acc); decentralized-AI-compute-viability (the centralized trillion-dollar buildout won 2024–26; no decentralized-inference breakout).
- Most surprising delta: he answered "concentration is dangerous" with his own de-concentration — engineered the EF restructure (2025-01), reduced himself to board oversight, and the foundation cut ~40% budget / ~20% staff through 2026 — revealed behavior almost no one else on the bench would replicate against their own institution.
- Employer/role check (mandatory): EF board member (with Miyaguchi, Storchenegger) after the 2025–26 restructure and departure wave (EF leadership tracker); no new commercial roles found. New falsifiable AGI definition (2026-07-20) and formal-verification security bet (2026-09-17) added with checks.
- Tier: quarterly — high essay velocity, live EF instability, and three active checks (coalition, decentralized-inference share, verification test case) that can move within quarters.
- Confidence unchanged at medium.
Sources
- Essays (primary, vitalik.eth.limo): "My techno-optimism" (2023-11-27) — the d/acc essay · "d/acc: one year later" (2025-01) · "From prediction markets to info finance" (2024-11-09) · "What do I think about Community Notes?" (2023-08-16) · "Crypto + AI applications" (2024-01-30)
- 2026 statements: CoinDesk — Ethereum's role in AI (X post, 2026-02-10) · Capability waves / AGI definition / 60-not-200 (X thread, 2026-07-20) · Decrypt — AI defense-favoring security, formal verification (2026-09-17)
- EF restructure: CoinDesk (2025-01-21) · EF leadership tracker (2025–26)
- Info-finance institutionalization: ICE strategic investment in Polymarket ($2B, 2025-10) · ICE +$600M (2026)
- Protocol: Fusaka upgrade (mainnet 2025-12-03) · Ethereum whitepaper (2013-11) · The Merge (2022-09-15)
- Books: Proof of Stake (ed. Nathan Schneider, 2022-09)
- Podcasts / talks: Bankless (multiple); Lex Fridman; Devcon keynotes (2014–); ETHCC
- X / Twitter: prolific primary source — lower tier per CONVENTIONS rule 4 but where his 2026 AI positions actually live
migrated v1→v2 2026-09-21 · backup: _archives/vitalik-buterin.html.bak-20260921 · converts-from: people/vitalik-buterin.md · AI & Society domain