Tristan Harris

persona · active · confidence: medium-high · reviewed 2026-10-03 · tier: quarterly

Who he is

Affiliation (role check run 2026-10-03): Co-Founder & President, Center for Humane Technology (CHT) — v1 said "Executive Director"; the ED is Julie Guirado, and Harris sits on the board alongside her (humanetech.com/team-board). Co-host, Your Undivided Attention (with Aza Raskin; TED Audio Collective; every other Thursday). Former Google Design Ethicist; studied under BJ Fogg at Stanford's Persuasive Technology Lab. Featured in The Social Dilemma (Netflix, 2020) and now a central figure in The AI Doc: Or How I Became an Apocaloptimist (Focus Features, theaters 2026-03-27; Netflix 2026-09-15). Compensation from CHT: $243,592 in FY2025 (Form 990 via ProPublica — a nonprofit's public tax return).

One-line position (v1, 2026-06): AI is the most powerful persuasion and manipulation technology ever built — the attention economy was the warm-up act — and without a fundamental shift in how AI systems are designed and governed we face an unprecedented assault on human agency, truth, and democratic institutions.

One-line position (what he actually argues now, 2025–26): "It's not ChatGPT that's the existential threat. It's the race to deploy … under the worst incentives possible" (Kara Swisher, 2026-03-26). A handful of labs are racing to build AI that replaces human labor; the race leaves no time to check whether the systems are controllable; the fix is a "narrow path" of liability, no AI personhood, pre-deployment testing, whistleblower protection, and a U.S.–China red phone — and the window is "12 to 24 months" (CNN Amanpour, 2026-04-08). The persuasion thesis is still in the file; it is no longer the thesis.

Discipline & technical bet

Discipline: persuasive-technology design turned advocacy. He is a non-technical thinker about a technical system — he does not train models, publish ML research, or hold a lab role — and he knows it: his method is to take other people's lab results (system cards, red-team findings, resignation threads) and translate them into public pressure. Model this as a frame-setter, not a forecaster.

Technical bet, now: two, both new since v1. (1) Humane Evals — CHT's proposal (2026-04-27, with Imran Khan) to measure what AI does to people (dependency, delusion reinforcement, critical-thinking erosion, crisis response) the way labs measure what AI can do; CHT positions itself as convener of "a small steering group," no numbers or funding yet (CHT, 2026-04-27). (2) Recursive self-improvement is the red line (AI building the next AI without people in the loop): he wants government or third-party pre-approval before any lab attempts it, and says some labs are "3–4 months away" (CHT, 2026-09-17). That is a falsifiable technical wager with a date; see Predictions.

Key claims (Says)

Notable predictions — with falsifiable checks

Revealed behavior (Does)

Feels

Fears being right too late — the Social Dilemma playbook (awareness first, law never) repeating at a scale where "the other issues won't matter" (Globe and Mail, 2026-09-24). Hopes for a Frances Haugen moment and believes he just got one: "This is the warning shot, and thank God it happened … a gift." Wants it to be true that clarity generates agency — that if heads of state saw what he sees, they would act.

Hears

Lab insiders and defectors first — system cards, red-team reports, the Coxon thread, Hubinger; then Bengio, Harari, Kokotajlo's timelines and Cotra's takeover assessments; the Vatican (he devoted an episode to Magnifica Humanitas, 2026-05-25). Little economics beyond Stanford's entry-level numbers until the Kinder episode; little from the open-source camp (Ng, Andreessen) except as the thing to argue against.

Sees

The one vantage he genuinely owns: the inside of a persuasion-design culture he was trained in, and a 2020-vintage public trust that lets him put lab findings in front of audiences (TMZ to Davos) no researcher reaches. Blind spot: he sees lab output through the labs' own red-team framing — the Opus 4 "blackmail" result he cites required engineered scenarios (Techdirt) — and he has no independent technical capacity to check it.

Incentive map

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

2026-10-03 — v1→v2 migration + validation (batch run, Fable)

Sources

migrated v1→v2 2026-10-03 · backup: _archives/tristan-harris.html.bak-20261003 · converts-from: people/tristan-harris.md · Singularity 🧠 / People