Yuval Noah Harari

persona · active · confidence: medium · reviewed 2026-09-21 · tier: semiannual

Who he is

Affiliation (role check run 2026-09-21): Historian and author of Sapiens (2014 in English), Homo Deus (2016), 21 Lessons for the 21st Century (2018) and Nexus: A Brief History of Information Networks from the Stone Age to AI (2024-09); "55 million copies in 65 languages" (his own bio — self-reported, not audited). The Hebrew University of Jerusalem history department still lists "Prof. Yuval Noah Harari" under Academic Faculty with no title, leave or departure note, but his own bio now says he "has been a lecturer at the Department of History" — past tense; treat Hebrew University as a legacy affiliation, not where his time goes. Co-founder (2019) of Sapienship, "a social impact company with projects in the fields of research, education and storytelling," with his husband and original agent Itzik Yahav (President); CEO Naama Avital; about a dozen named leadership staff including a Head of Rights and a Director of Chinese Speaking Markets. Distinguished Research Fellow at Cambridge's Centre for the Study of Existential Risk (per his bio; CSER's page returned 403 this cycle, not verified on the institution's side). New this cycle: a Director on the Board of LawZero, Bengio's safe-by-design AI nonprofit — appointed 2026-01-15 ("The appointees to the Board of Directors are: Maria Eitel (Chair of the Board), Mariano-Florentino (Tino) Cuéllar, Yuval Noah Harari, Valérie Pisano, Sir John Rose, Yoshua Bengio"), confirmed on LawZero's own team page as "Director, Board" and in his bio ("In 2026 he joined the AI safety organization LawZero, as a Director on its board"). On 2026-09-16 LawZero announced that "Canada and Germany announced funding of up to CAD $300 million for LawZero." That seat is the affiliation change the June-2026 v1 review could not have seen; it is also the first time the bench's leading AI narrator sits inside an organisation that builds AI.
One-line position (v1, 2024): AI is the most powerful information network ever created, and like every prior information network it will reshape power, truth and human self-understanding — except this time we may lose the ability to correct course because AI doesn't know what it doesn't know.
One-line position (2026): AI is not a tool but an agent — "alien intelligence" — and the danger is not a robot uprising but a voluntary handover: people will cede finance, information and even intimacy to AIs because they are more efficient and more trusted than other humans. "Trust will migrate from humans to AIs" (The Economist, 2026-08). The 2024 thesis about networks became a 2026 thesis about trust; the file grades both.

Discipline & technical bet

Macro-historian (big-picture history across thousands of years) with no technical wager of his own: he writes no code, funds no lab, publishes no evals. The closest thing to a bet is institutional and policy-shaped. Institutional: the LawZero board seat (2026-01) — a lab whose product thesis is non-agentic AI, which is the engineering form of his "agents are the danger" argument; joining it is the one place his behaviour puts a stake behind the talk. Policy: three concrete asks, all from 2026 — "A ban on AI personhood"; "A ban on AI mimicking, impersonating, pretending to be human beings or pretending to be conscious"; and "Companies should not train their AIs to deliberately impersonate humans, to deliberately pretend to have feelings" (The Economist, 2026-08). Where his time went in the window: on his own article index, 14 bylined essays or long interviews between 2025-01-03 and 2026-08-23, about nine of them AI-centred (Wired ×2, The Japan News, WSJ, FT, El País, The Economist, NYT/Klein in part) and five on Israel–Palestine, Ukraine or Trump; a main-stage Davos keynote (2026-01-20, his third after 2018 and 2020); the India podcast circuit (Nikhil Kamath, 2026-02-11); and a role in the 2026 documentary The AI Doc. Output cadence is up, not down — a commentator's "job change" is a book or a platform, and this cycle the platform grew.

Key claims (Says)

Notable predictions — with falsifiable checks

Revealed behavior (Does)

Feels

Fear: that the handover is voluntary — people will hand finance, media and intimacy to AIs "because they are more efficient" and nobody will need to seize anything. Hope, stated plainly to an Indian audience: "Democracy is not dead. It's still extremely powerful. It's still the most powerful idea around" (2026-02-11). What he wants to be true: that story-telling primates can still choose their stories. Underneath: an Israeli who believes his own country is failing its self-correction test — the Netanyahu argument and the AI argument are one argument.

Hears

Bengio (now a fellow board member), the Cambridge existential-risk community, Davos and Economist editors, Sapienship's research staff, and long meditation retreats. Largely absent: working ML researchers, labour economists (his jobs claims never cite payroll data), and the Chinese readers who buy him.

Sees

A vantage nobody else on the bench has: 55 million books sold plus a Davos main stage plus a safe-AI lab boardroom — he sees how elites receive the AI story better than anyone. What he does not see: code, evals, data pipelines or payrolls; his mechanisms are inferred from history, which is why the dated ones (floods, coders) are the ones that get graded.

Incentive map

Re-derived this cycle. Who pays: readers (55 million copies, self-reported); Sapienship (his and his husband's company, which owns and sells his IP); keynote and speaker-bureau fees (bureau-listed; fees not public — third-party "net worth" sites are not sources and are excluded); a Cambridge CSER fellowship (terms unknown); and, from 2026-01, a LawZero board seat (compensation unknown — no Canadian nonprofit return read this cycle). What he's selling: one story — AI is an agent, the fix is restraint — and it sells books, Davos slots and now a lab whose product is non-agentic AI with up to CAD $300M of public backing. Newest vehicle matches newest thesis: the personhood and impersonation bans (2026) are the policy shape of LawZero's technical thesis; log the pairing. What he can't say: that AI risk is overstated; that the useless class is behind schedule; anything that undercuts LawZero's public-funding case; and — a test to keep running — anything China-critical enough to cost the Chinese market. Household: manager and company President are the same person; there is no arm's length between the brand and the family. Independence decay: Harari and Bengio now sit on the same board — the bench can no longer count them as two independent voices on AI risk. Rule-25 discount (CONVENTIONS): high on timelines ("no human money in 10 years," 2028 thresholds), moderate on historical claims, and a new institutional discount on any LawZero-adjacent policy ask.

Theories aligned with

What he's reacting against

Where he overlaps / splits (with Rich)

Track record

Empirical vs normative

Weak spots / open questions

Rich's take

Delta log

2026-09-21 — v1→v2 migration + validation (bulk batch)

Sources

migrated v1→v2 2026-09-21 · backup: _archives/yuval-noah-harari.html.bak-20260921 · converts-from: people/yuval-noah-harari.md · AI & Society domain