AI world · US AI Safety Institute● safety ecosystem
Paul Christiano
Head of Safety
About
Invented RLHF, founded ARC, now leads safety at the US AI Safety Institute; the field's canonical alignment researcher.
Media coverage indexunvetted
Sources & placements — spoke to them or published them
Lexi Pandell — Wired
- Self-Driving Cars. Rogue Nuke Launches. Evil AI. What Tech Threats You Should (and Shouldn't) Worry Aboutsource · direct quote sympathetic · Aug 1, 2017 — Quotes Christiano on AI eventually doing research/construction and divergence from human goals.
The Economist staff — The Economist
- How generative models could go wrongsource · direct quote neutral · Apr 19, 2023 — Quotes Christiano on ARC considering AI-safety industry standard and lab reactions.
TIME staff — TIME
- Paul Christiano: The 100 Most Influential People in AI 2023source · interview · Sep 7, 2023 — TIME sat down with Christiano directly ('TIME spoke with Christiano...condensed and edited for clarity') about inventing RLHF, leaving OpenAI, and ARC's work.
Will Henshall — TIME
- Paul Christiano — TIME100 AIsource · interview sympathetic · Sep 7, 2023 — Profile/interview on RLHF development, leaving OpenAI, ARC evaluations, outside scrutiny of labs.
Asya Bergal — AI Impacts
- Paul Christiano on the safety of future AI systemssource · interview sympathetic · Sep 11, 2019 — Detailed discussion of AI risk, tractability of safety work, takeoff, alignment progress.
Bankless hosts (Ryan Sean Adams / David Hoffman) — Bankless
- Why AI Alignment Matterssource · interview · Apr 27, 2023 — Hosted long-form interview in which Christiano gave his widely-cited '50/50 chance of doom' estimate, later re-reported by other outlets (Fortune, Business Insider, Yahoo).
Daniel Filan — AXRP (AI X-risk Research Podcast)
- Episode 12 - AI Existential Risk with Paul Christianosource · interview · Dec 2, 2021 — Hosted long-form interview specifically on AI existential risk research.
Dwarkesh Patel — The Lunar Society / Dwarkesh Podcast
- Paul Christiano - Preventing AI Takeoversource · interview · Oct 31, 2023 — Long-form hosted interview covering RLHF invention, AI timelines, responsible scaling policies, and ARC's proof-system research.
- Paul Christiano — Preventing an AI takeoversource · interview sympathetic · Oct 30, 2023 — ~3hr interview on takeover risk, gradual vs fast transitions, scalable oversight, governance.
Future of Life Institute staff — Future of Life Institute
- AI Researcher Paul Christianosource · interview sympathetic · Sep 30, 2016 — Researcher profile/interview on route into AI safety and alignment priorities.
Katja Grace — AI Impacts
- Conversation with Paul Christianosource · interview · Aug 1, 2019 — AI Impacts research-conversation writeup/interview on alignment doominess estimates and timelines.
Rob Wiblin — 80,000 Hours Podcast
- Paul Christiano on how OpenAI is developing real solutions to the 'AI alignment problem'source · interview · Apr 6, 2021 — Long-form hosted interview on OpenAI's alignment approach and Christiano's vision for AI decision-making handover.
- Paul Christiano on whether we should leave a helpful message for future civilisationssource · interview · Oct 2, 2019 — Separate hosted 80,000 Hours episode, distinct from the 2021 alignment-solutions episode.
Robert Miles — The Inside View (podcast)
- Paul Christiano's views on "doom" (ft. Robert Miles)source · interview · Sep 29, 2023 — Interview/discussion episode featuring Christiano's own writeup of his p(doom) views, discussed live with Robert Miles.
Ryan Sean Adams — Bankless
- 168 — How to Solve AI Alignment with Paul Christianosource · interview sympathetic · Apr 22, 2023 — ~2hr interview on takeover probabilities, timelines, scalable oversight, honesty, coordination.
Mentions — wrote about them
Andrew Marantz — The New Yorker
- Among the A.I. Doomsayersmention · secondhand neutral · Mar 11, 2024 — Describes AI Impacts role, relationship with Katja Grace, OpenAI tenure, ARC, rationalist/EA links.
Anthony Cuthbertson — The Independent
- ChatGPT creator says there's 50% chance AI ends in 'doom'mention · secondhand neutral · May 3, 2023 — Overbroadly describes him as ChatGPT creator; highlights 50/50 doom estimate.
Ashley Belanger — Ars Technica
- Feds appoint 'AI doomer' to run AI safety at US institutemention · secondhand neutral · Apr 17, 2024 — Extended account of RLHF work, risk estimates, appointment controversy, evaluations views, pause position. [Valence re-judged under seriousness axis: Extended, substantive account of his RLHF work, risk estimates, and the appointment controversy.]
Ben Paynter — Fast Company
- Take A Chance With Your Charity And Try A Donor Lotterymention · secondhand sympathetic · Jan 31, 2017 — Mentions Christiano in connection with donor-lottery idea.
Benj Edwards — Ars Technica
- OpenAI checked to see whether GPT-4 could take over the worldmention · secondhand neutral · Mar 15, 2023 — Describes ARC's GPT-4 autonomy tests and defense of evaluating dangerous capabilities.
Bernard Marr — Forbes
- Should We Stop Developing AI For The Good Of Humanity?mention · secondhand neutral · May 3, 2023 — Cites Christiano's 10-20% takeover-and-mass-death estimate.
Billy Perrigo — TIME
- How Commerce Secretary Gina Raimondo Became America's Point Woman on AImention · secondhand neutral · Jun 6, 2024 — Notes appointment and grave AI-risk forecasts; quotes Raimondo defending value of his views inside NIST.
Cade Metz — The New York Times
- Teaching A.I. Systems to Behave Themselvesmention · secondhand · Aug 13, 2017 — Profiles Christiano's RLHF work with Dario Amodei/DeepMind; describes the research rather than a fresh quote obtained from Christiano for the piece.
- Teaching A.I. Systems to Behave Themselvesmention · secondhand sympathetic · Aug 13, 2017 — Reports on Christiano and Dario Amodei's human-preference RL work as notable safety advance.
Cat Zakrzewski — The Washington Post (Technology 202)
- Top tech companies seek more cash for the lab keeping AI safemention · passing mention neutral · Apr 23, 2024 — Reports appointment as head of AI safety and 20% risk estimate.
Dylan Matthews — Vox
- The $1 billion gamble to ensure AI doesn't destroy humanitymention · secondhand neutral · Sep 25, 2023 — Discusses Christiano as Anthropic Long-Term Benefit Trust member and safety-network figure.
Jeremy Kahn — Fortune (Brainstorm AI newsletter)
- Fortune Brainstorm AI London is gathering top AI decision makersmention · secondhand neutral · Mar 12, 2024 — Summarizes VentureBeat report of internal NIST opposition to Christiano's EA/x-risk orientation.
John Werner — Forbes
- AI Safety Group Finds Agentic Tools Slow Down Open Source Devsmention · passing mention neutral · Sep 28, 2025 — Background mention identifying ARC as nonprofit Christiano founded after OpenAI alignment team.
Kelsey Piper — Asterisk Magazine
- A Field Guide to AI Safetymention · secondhand · Jun 1, 2023 — Field guide to the safety landscape; identifies Christiano as ARC head and RLHF co-developer, no fresh quotes. Hand-added after a human reader flagged the scan miss.
- How to test what an AI model can — and shouldn't — domention · secondhand sympathetic · Mar 29, 2023 — Explains ARC's GPT-4 evaluations and Christiano's founding role.
Kyle Wiggers — TechCrunch
- NIST launches a new platform to assess generative AImention · secondhand neutral · Apr 29, 2024 — Calls appointment controversial due to 'doomerist' views; references NIST scientist concern.
Sharon Goldman — VentureBeat
- NIST staffers revolt against expected appointment of 'effective altruist' AI researcher to US AI Safety Institutemention · secondhand neutral · Mar 8, 2024 — Reports staff resignation threats over EA/longtermist associations; also quotes defense of qualifications. [Valence re-judged under seriousness axis: Balanced reporting on internal NIST backlash that also includes a defense of his qualifications.]
Tristan Bove — Fortune
- OpenAI's former top safety researcher says there's a '10 to 20% chance' that the tech will take overmention · secondhand neutral · May 3, 2023 — Reports Bankless estimates and departure from OpenAI to ARC.
Unknown (Axios Pro tech policy team) — Axios
- Exclusive: Biden administration taps top AI talentmention · secondhand · Apr 16, 2024 — Axios exclusive announcing Christiano's appointment as AISI Head of AI Safety, citing Susan Rice's praise of him; article was paywalled (403) so could not confirm if Christiano himself was quoted directly — treating as secondhand pending verification.
Unknown (VentureBeat staff) — VentureBeat
- NIST staffers revolt against expected appointment of 'effective altruist' AI researcher to US AI Safety Institutemention · secondhand · Apr 19, 2024 — Covers internal NIST pushback over Christiano's appointment as AISI Head of AI Safety; no direct quote from Christiano identified in coverage.
Beth Barnes — METR/ARC Evals blog
- ARC Evals is spinning out from ARCmention · secondhand sympathetic · Sep 19, 2023 — Explains Christiano founded/led ARC and would remain head as evaluations team spun out.
Chloe Xiang — Vice/Motherboard
- GPT-4 Hired Unwitting TaskRabbit Worker By Pretending to Be 'Vision-Impaired' Humanmention · passing mention neutral · Mar 15, 2023 — Identifies Christiano as ARC leader and former OpenAI alignment team head.
Darren Allan — TechRadar
- Will AI spell doom for humanity? One of ChatGPT's creators thinks there's a 50% chancemention · secondhand neutral · May 4, 2023 — Repeats 50/50 doom statement, contrasts with Yudkowsky's faster takeoff view.
José Antonio Lanz — Decrypt
- Former OpenAI Researcher: There's a 50% Chance AI Ends in 'Catastrophe'mention · secondhand neutral · Apr 28, 2023 — Reports Bankless statements on takeover and catastrophe risk.
Lucas Nolan — Breitbart News
- Former OpenAI Researcher: AI at Human Level Means a '50/50 Chance of Doom'mention · secondhand neutral · May 4, 2023 — Recycles Bankless coverage, compares position with Yudkowsky's.
NIST staff (unnamed) — U.S. Department of Commerce/NIST
- U.S. Commerce Secretary Gina Raimondo Announces Expansion of U.S. AI Safety Institute Leadership Teammention · passing mention sympathetic · Apr 16, 2024 — Announces Christiano as head of AI safety responsible for frontier-model evaluations.
Robert Wiblin — 80,000 Hours Podcast
- Brian Christian on the alignment problemmention · passing mention neutral · Feb 1, 2021 — Includes Christiano among 'Concrete Problems in AI Safety' authors.
- Vitalik Buterin on effective altruism, better ways to fund public goodsmention · passing mention neutral · Jan 1, 2019 — Wiblin asks Buterin whether he knows Christiano.
Rohin Shah — AI Alignment Newsletter
- Alignment Newsletter #33mention · secondhand sympathetic · Dec 1, 2018 — Detailed treatment of Christiano's clarification of 'AI alignment.'
- Alignment Newsletter #27mention · secondhand sympathetic · Oct 9, 2018 — Summarizes the four-hour 80,000 Hours interview and Christiano's alignment strategy views.
- The Alignment Newsletter #2: 04/16/18mention · secondhand sympathetic · Apr 16, 2018 — Summarizes Christiano's definitions of alignment and proposed guarantees.
Ryan Sean Adams — Bankless
- DEBRIEF — How to Solve AI Alignmentmention · secondhand sympathetic · Apr 22, 2023 — Follow-up discussion reflecting on implications of Christiano interview.
Sebastian Klovig Skelton — Computer Weekly
- Government AI taskforce appoints new advisory board membersmention · passing mention neutral · Sep 7, 2023 — Reports appointment to UK Frontier AI Taskforce advisory board.
Misc — low-signal outlets (8 mentions)
AI Weekly staff — AI Weekly
- Artificial Intelligence News #67mention · passing mention neutral · Aug 18, 2017 — Highlights and links NYT report on Christiano's work.
AI-SCHOLAR staff — AI-SCHOLAR
- RLHF: How to Train Reinforcement Learning Agents Using Human Evaluationmention · secondhand sympathetic · Jan 9, 2023 — Technical explainer crediting Christiano's preference-learning method as RLHF basis.
Amelia Jones — UNILAD
- Expert warns there's a 50% chance AI could end in 'catastrophe' with 'most' humans deadmention · secondhand neutral · May 18, 2023 — Repackages Bankless estimates as warning of 50% catastrophe risk.
EPIC for America staff — EPIC for America
- How the Biden-Harris Administration Created an Unlawful AI Regulatory Agencymention · secondhand dismissive · Oct 1, 2024 — Calls Christiano's 10-20% AI-takeover estimate 'apocalyptic thinking' and questions his suitability for AISI. [Valence re-judged under seriousness axis: Labels his estimate 'apocalyptic thinking' and questions his suitability rather than engaging the substance.]
Jeremi Levesque — Jeremi's Blog
- Deep reinforcement learning from human preferencesmention · secondhand sympathetic · Feb 2, 2024 — Reviews Christiano et al.'s foundational human-preference RL paper.
Reflections on AI staff — Reflections on AI
- Bankless Interview with Paul Christianomention · secondhand sympathetic · May 2, 2023 — Summarizes interview; notes Yudkowsky recommended Christiano as less pessimistic voice.
TFTC staff — TFTC
- 'AI Doomer' Paul Christiano Appointed to Lead US AI Safety Effortsmention · secondhand neutral · Apr 1, 2024 — Covers appointment foregrounding high catastrophic-risk estimates. [Valence re-judged under seriousness axis: Confirmed via fetch: straightforward reporting of the appointment and risk estimates, including both critics and supporters.]
Tom Rochette — Tom Rochette's Blog
- Paul Christiano — Deep Reinforcement Learning from Human Preferences (2017)mention · secondhand neutral · Jan 1, 2017 — Technical notes summarizing the preference-learning paper.
Scan coverage note: Report covers July 15, 2016-July 15, 2026 via searches of name variants, affiliations, technical terms, and niche AI-safety sources; excludes academic papers, Christiano's own posts, ordinary user comments, and database pages. Stated gaps: no access to subscription press/broadcast databases (LexisNexis, Factiva, ProQuest, TVEyes, Critical Mention, Meltwater, Cision); NYT, Bloomberg, BBC, and MIT Technology Review indexes were only partially crawlable; X/Twitter historical archive, deleted posts, private accounts, Facebook, LinkedIn's full archive, and closed Slack/Discord communities could not be comprehensively searched; podcast platforms lack full transcript-level search so directories/indexed transcripts were used instead; wire-service syndication duplicates were not exhaustively listed (only identifiable outlet rewrites retained); no substantive standalone journalist social-media thread about Christiano was found beyond linked articles. [Merged: codex deep harvest 2026-07-15 + earlier native scan + hand adds.]