Updated: 2026-11-16
What he is here for: CEO of OpenAI, the most-cited AI executive on Moonshots with Peter Diamandis without a page (20 of 101 indexed episodes mention him; ROLLUP promotion candidate). Promoted 2026-10-08 off Moonshots #301, which gives him a segment of his own.
Why it is in this archive
Gate: cited in 20 episodes. Only the two episodes above were read for this page; the other 18 are not read. Nothing here says "he never says X".
The line (Politico Decoded, 4 Oct 2026)
Tape (0:40:16, Peter reading): "We believe that the world should accept some bad things happening for the benefit of technology that people will obtain, that people should have agency." Politico's own index dates the piece 4 Oct 2026, 8:55 PM UTC, Christine Mui and Brendan Bordelon, headline "Sam Altman to Decoded: 'The world should accept some bad things happening' for the benefits of AI" (index metadata only; the article itself is not opened, the fetch is blocked). Secondary accounts carry the fuller sentence: "One of the differences between us and some of the stricter AI safety people is that we believe that the world should accept some bad things happening for the benefits of this technology and people having the agency" (Taipei Times editorial 13 Oct; Iowa AG's wording 14 Oct matches). Coverage agrees (ExplainX, secondary). Also reported, and not read on air: he did not extend it to catastrophic risk, including a serious loss of control of AI; he described a "lighter touch" against Anthropic's stronger regulation; and, same interview, a "coming tidal wave of cybersecurity problems" from open models, with society having to "accept some fairly severe cyber incidents from open models, in exchange for the liberty that comes with that" (The Hill via Yahoo, 13 Oct; secondary).
Three weeks earlier
Techmeme, from his X post of 12–13 Sep: he agreed with Amodei that independent evaluators with employee-like access are a great idea and said OpenAI would match it. On tape Alex puts different words in his mouth: "Sam's response was, 'I'm in. Sign me up. Open AI wants to be part of this cartel.'" (0:43:37): no source found with those words; "cartel" is Alex's own term from Moonshots #291. Alex's reading now: "that plan is dead. The cartel is over" (0:43:53).
What the panel said (not facts)
- Dave: "It's about taking the moral high ground and winning the PR war… I don't think Sam can guarantee safety… it's actually a really smart message" (
0:41:59–0:43:07). - Alex: bad things are "the cost of freedom and the cost of competition… the cost of progress" (
0:44:07). - Salim: "the goal should not be zero bad outcomes, that means you just ban the technology" (
0:46:22); FARs "written in blood" (0:46:04). - Emad: a "mapping project" of the optimist and the scared scenarios (
0:49:48). - OpenAI president Greg Brockman, not Altman, is the accord signatory (29 Sep; secondary).
This week (checked 2026-11-16, window 10–16 Nov)
One new appearance on record: a 30-minute stage interview at Semafor's "Silicon Valley & The World" on Mon 16 Nov, read at Semafor and TechCrunch. It was his first public comment on Jensen Huang's exit from Nvidia. He repeated the no-handover line, said OpenAI could not safely push recursive self-improvement "right now", and backed outside evaluators in principle. Missed last pass: OpenAI's CFO on CNBC (6 Nov) said the round is priced pre-money, which settles the pre/post question as far as OpenAI's own word goes. No closing announcement. Foundation unchanged (newest item still 14 Oct). No new open call.
- Mon 16 Nov, Semafor "Silicon Valley & The World", interviewed by Reed Albergotti. Semafor, J.D. Capelouto, updated 16 Nov 4:14 PM EST (read; the host outlet's own account, the closest to primary this desk has). TechCrunch, Rebecca Bellan, 16 Nov 10:11 AM PST (read). Venue: Rosewood Sand Hill, per Seoul Economic Daily only (an AI translation from Korean that flags its own quotes as not exact; not used for wording). Recording not watched. Quotes:
- Asked whether Huang should return to Nvidia: "I don't think I know enough to answer that question." Both outlets agree on this one.
- On Huang (TechCrunch): "It's hard to imagine the AI revolution without Jensen and the company he has built"; "Nvidia is, like, a behemoth of a company… I think people still trust their guarantees very much"; "the revenues and demand are growing so strongly that I think the industry will be fine… Jensen has played this extremely special and valuable role as someone to really drive things forward with great ambition." Semafor quotes "long been a hero of mine. Heroes have flaws" and "personal strong empathy". Wccftech's longer transcript (16 Nov, read) differs in wording, so this page uses only what Semafor and TechCrunch both carry. The empathy line refers back to his own 2023 removal (Wccftech: "I remember what it was like").
- On control, the same line as ADIPEC on 2 Nov, in new words (Semafor): "We should not want to have a 'handover' … either gradual or sudden, to AI making decisions about our future."
- On recursive self-improvement (Semafor): OpenAI could push toward it, but "I don't think we could safely do that right now." This is a new on-record line for this page.
- On outside evaluators (Semafor, paraphrase with a quoted fragment): the METR principle "is going to be increasingly important and valuable." He named no contract (see Open).
- Also in Semafor: "We had the chatbots, then we had the coding agents. I think in the last few months we started to see the beginning of AI for science"; OpenAI "will go public at some point", with "real downsides to being public"; on raising money, "That is not a worry… There's been way too much investor demand and not enough supply"; on data centres, "We have a giant country, we have a lot of space we can do this."
- Numbers: the only figure in the coverage is TechCrunch's own: Nvidia agreed to guarantee up to $105B backing OpenAI's Ohio data-centre project. That is TechCrunch's statement, not his, and the guarantee document has not been read.
- Missed last pass, Fri 6 Nov: the CFO on the round. CNBC's "unofficial transcript" of Sarah Friar on Squawk on the Street (Pebble Beach), published 6 Nov 12:43 PM EST, read: "We will raise at a $1.25 billion pre-money valuation. The round is ongoing, but we feel quite confident in it." The transcript says "billion"; Eisen's question and all coverage say trillion, so this is read as a transcription slip (video not watched). She also said "1.25, remember, is almost a 50 percent lift in valuation from the round we closed in March". On going public: "I think last we said we would continue to look in '27." She gave no size for the round. This answers the pre/post-money question on OpenAI's own word. The March figure was post-money, so her "50 percent" compares a pre-money with a post-money number.
- The round: still not closed in public. A "$15B at $1.25T" figure, attributed to CNBC, turns up in a Benzinga URL (403, not read), an MT Newswires sidebar headline and a 13–14 Nov aggregator. CNBC's own article was not found, so the number is not carried. OpenAI has announced nothing.
- Moonshots: EP310 (11 Nov) does not mention him; one "Sam Alliance" at
1:14:30is a garbled caption, speaker and meaning unresolved. EP311 (13 Nov) has panel talk only, with no tape of him:0:16:53compares Huang to "Sam step out of OpenAI for just a little bit and being brought back", and0:52:36has "If I were Sam", the advice being to announce OpenAI solving "the top 5,000 diseases" (speaker not resolved). At1:19:51a panelist says he "just saw a headline that open AI has disbanded their robotics team". Not verified: search finds only the 2021 disbanding, and it would conflict with his 29 Oct "robotics program… data centers" line (EP307, below). Do not repeat it. - OpenAI product items (company, not him; secondary, openai.com 403): ChatGPT Wallet (biometric approval per purchase; Amex, Mastercard and Visa; US Pro subscribers first; The Neuron 16 Nov) and a child-safety image classifier announced at a Washington press briefing on 16 Nov (one outlet, citing The Verge; not read). Not carried: "Disney's $1B partnership fell apart after Sora's shutdown" (The Neuron, one line, no source).
- Assessors: still no contract announced by 16 Nov (search). His Semafor line backs the principle, but the 9 Oct "finalizing contracts" statement is now five weeks old.
- Foundation unchanged: openaifoundation.org news (both pages, 15 items) and home read 16 Nov. The newest item is still 14 Oct (Cities Deliver), and no new open call is listed (the Aug 20 hiring post lists roles, not a programme). Cities Deliver Round 1 is still due 29 Nov 2026 (already in
submissions/pipeline.jsonasprize-cities-deliver-2026). Not checked this pass: openai.com (403), his X timeline (402), Bloomberg (403), the state-suit dockets.
Previous week (checked 2026-11-10, window 4–10 Nov), kept
No new appearance by him dated 4–10 Nov was found. The week's one hard fact is money: a reported first close of about $11B at $1.25T (Bloomberg, not read). Four items from 31 Oct–3 Nov were missed last pass and are added here: the Korea visit, the Figuring Out podcast, the "people should always be in control" X post, and the ADNOC–OpenAI–G42 partnership at ADIPEC. The RAM episode ended with his display name restored. Foundation unchanged. No new open call.
- Fri 6 Nov: first close of the round (reported). Bloomberg, 6 Nov, "OpenAI Completes $11 Billion Portion of Funding at $1.25 Trillion Value" (not read; carried by Let's Data Science 6 Nov and Nairametrics 7 Nov, both read, both citing Bloomberg's unnamed people familiar). About $11B had been wired "as of Friday", at a $1.25T valuation, and the round could reach $20B in total. Coatue and T. Rowe Price are named in the first close. Earlier talks referred to about $1.4T (per Bloomberg, as carried). OpenAI declined to comment and has announced nothing. Still unsettled: pre- or post-money (TechCrunch 22 Oct said pre-money; CNBC/FT 31 Oct said post-money). He is not quoted. Not used: a Dealroom "final $20B from Nvidia and SoftBank" item turned up in search; it is dated 1 Oct and concerns the March round, not this one.
- RAM episode, closed. TechJuice (no absolute date, read; secondary) reports that his X display name went back from "Ram Altman" by 11 AM ET on 6 Nov, and quotes his Wed 4 Nov replies: he has "long pushed for the world to make more memory", "i want you to have so much ram lets build more", and, of the rename, "i lol'd". X itself was not opened (402). A developer's "Ram-Altman" email script was covered by Cybernews on 9 Nov (403, headline only).
- Missed last pass, 3–4 Nov: human-control X post. ANI via Devdiscourse, 4 Nov 10:10 IST (read; syndicated, unedited), so the post went up on or before the evening of 3 Nov ET: "People should always be in control of the future" and "people should not cede control of the future to superintelligence, and that requirement should ultimately constrain AI progress". ANI also paraphrases a rejection of humans being "relegated to performing tasks that AI systems consider worthy of reward" (paraphrase, not his words). One outlet in search says the wording was borrowed, without saying from whom: not verified. The post itself was not opened (x.com 402). It repeats his ADIPEC line of 2 Nov ("should not be a handoff to AI").
- Missed last pass, published 3 Nov: Figuring Out with Raj Shamani, episode FO583 (YouTube
TjPPIf-M6KE; recording not watched). R&D World, 5 Nov, Brian Buntz (read; secondary) quotes him: "The math results were quite amazing to me and it really made me think that we are closer to this milestone than I expected us to be", the milestone being a machine for scientific discovery, and "This is like the real deal" about the period before the September math release. The same piece credits him with "order $10 million" as the cost of the proof, from the 29 Oct Cheeky Pint tape. These are his claims. OpenAI's math repository is reported at 722 manuscripts, with 3 withdrawn and 14 revised on 7 Oct (R&D World's account; repository not read). Search summaries say the episode also covers the Sutskever and Murati departures and India's ChatGPT user base; not read at source. - Missed last pass, Sat 31 Oct: unannounced South Korea visit. CNBC, Sun 1 Nov 9:58 PM EST, Jenny Lee (read): per Samsung, he met around 50 executives from Samsung affiliates in Yongin on Saturday. Per Hankyung, as carried by CNBC, he had lunch with SK Group chairman Chey Tae-won, with SK Hynix and SK Telecom executives present. SK Hynix could not confirm the meeting. He left the same day, per local media. He is not quoted.
- Missed last pass, 2 Nov: the ADIPEC partnership. Fox News, 6 Nov 5:00 AM EST, Efrat Lachter (read in part; the second half was a model summary): at the ADIPEC opening he joined UAE officials to announce an ADNOC–OpenAI–G42 partnership to bring AI into the energy industry. The same day, per a photo caption, he and Mark Chen announced a "Machine-Readable Government" initiative with Abu Dhabi's crown prince. The claim that the partnership cuts development planning "from months to hours" is ADNOC's, and the article flags it as not independently verified. No terms or dollar figure were given.
- Moonshots: EP308 (4 Nov; recorded 3 Nov) brings him up three times, all as panel talk with no tape of him. In the cold open the RAM remark becomes "Sam's dream" (
0:08:32–0:10:52). At1:39:54a panelist hopes that "Mark Zuckerberg and Sam Altman will reserve some" compute for health. At1:47:44comes "Sam's hypothesis, robots that build data centers" (the EP307 Cheeky Pint line). ⚠️ EP308's captions are a machine round-trip translation, so none of these wordings is safe to quote without the tape, and speakers are not resolved. EP309 (7 Nov) does not mention him. - Not used: an emergent.sh piece (8 Nov) says he "revealed why top OpenAI talent left". It gives no source or occasion, and its departure dates conflict with the public record. Also not used: a 9 Nov repost of his "Team Humanity" X post, which dates from 14 Sep, not this week.
- Assessors: still nothing. No third-party assessor contract announced by 10 Nov (search). The 9 Oct "finalizing contracts" line is now a month old.
- Foundation unchanged: openaifoundation.org news (both pages, 15 items) and home read 10 Nov. The newest item is still 14 Oct (Cities Deliver), and no open call or deadline is listed. Cities Deliver Round 1 is still due 29 Nov 2026 (already in
submissions/pipeline.jsonasprize-cities-deliver-2026). Not checked this pass: openai.com (403), his X timeline (402), Bloomberg (not opened), the state-suit dockets.
Earlier (checked 2026-11-04, window 29 Oct – 4 Nov), kept
One major appearance, read at the outlet that held the interview (ADIPEC, 2 Nov, carrying a Millennium Prize claim that is his and not verified), a backlash cycle off the 29 Oct Cheeky Pint tape with a reported X walk-back, and the funding round cut to $20B — still talks. Foundation unchanged. No new open call.
- Mon 2 Nov, ADIPEC opening ceremony, Abu Dhabi, in conversation with The National's editor-in-chief Mina Al-Oraibi. The National, 2 Nov (updated 4 Nov 6:24 AM), Jennifer Gnana, read — the interviewing outlet's own account, the closest to primary this desk has. On human control of AI: "There are a lot of other things we can debate, but on this one, we think there must be no room for debate"; there "should not be a handoff to AI" (fragment); "I can't believe that this differentiates us from other AI companies, but it does. This is a value that we intend to really go fight for. And I wish it didn't need to be said." On capability: "AI was mostly not as smart as our smartest scientists, and now it's mostly smarter than them"; AI can now do "a lot of parts" of many white-collar jobs; society would "navigate through this all fine" (fragments). On safety: "How do we guarantee, and how do we give the world comfort with that guarantee, that we are going to avoid serious safety and alignment incidents? … I think that'll be increasingly important for our industry to demonstrate." Quick-fire, biggest challenge today: "Energy." Costs fall "in almost every way, except, you know, atoms are still atoms"; power is "the very dominant factor" in how much AI the world can use. His numbers on stage, his claims, not verified: OpenAI's heaviest internal user six years ago used ~100,000 tokens/month — today that is "roughly the global per-capita average" and the top user is "about 500 billion" (unit not stated in the article; tokens implied); ~80% of young people in the UAE use ChatGPT (article's paraphrase).
- The Millennium Prize line (same session), his claim, not verified. The National's text: OpenAI "had solved one of the Millennium Prize problems", using 10,000 agents running an upcoming GPT model for about four days, collaborating with "no hierarchy" (the only quoted fragment); the article itself says the claim "could not immediately be verified independently". Search-level context, none of it read at source this pass: this restates a claimed Navier–Stokes result OpenAI announced around early September (Axios URL dated 8 Sep 2026: "historic math solution overshadowed by credit controversy"; mathematician Tristan Buckmaster reported alleging the work tracked his, Altman rejecting it; The Decoder on the dispute; OpenAI reported saying it will not claim the $1M prize and that it "cannot rule out" de-identified rival-product data helped its models). Accounts conflict — proof vs proposed counterexample, "88 hours" vs "about four days". The Clay Mathematics Institute considers solutions only ≥2 years after publication; nothing is recognized. Do not repeat as fact; "Altman says" only.
- RAM backlash off the Cheeky Pint tape (29 Oct, the same episode EP307 clipped — see below), walk-back reported 3–4 Nov; all secondary, X not opened (402). On the tape he called consumer DRAM allocation "quite terrible" — "quite terrible to have everybody have a bunch of the world's limited supply" of DRAM sitting in PCs, used about "5% of the time" (his figure). Coverage wave 30 Oct–1 Nov (Tom's Hardware, TechRadar, Fortune via Yahoo — Fortune's "could soak up as much as 40% of supply" framing is Fortune's number, not his). Kotaku (4 Nov) and Futurism via Yahoo report his X responses: the remarks were misread, "I actually do want the world to have tons of memory", "I have been arguing for the world to ramp up memory production for a long time", and after a joke exchange he agreed to rename himself "Ram Altman" (outlets put the posts at 3–4 Nov; the posts themselves were not opened, dates not verified).
- Sat 31 Oct — funding round cut. CNBC, 31 Oct 3:56 PM EDT, Ashley Capoot and Lora Kolodny, read: OpenAI is now seeking up to $20B, "paring back its earlier target of up to $30 billion"; the FT (named and linked, not read) first reported the $10B cut, a $1.25T post-money target and a Friday deadline for investors to commit, "with some investors opting to reduce their allocations"; UAE investors lead alongside Coatue (CNBC's own 22 Oct reporting, not re-read); the round "is still in progress and has not closed". CNBC's basis for "not closed" is one unnamed person familiar with the deal; OpenAI did not respond. He is not quoted — the article only captions his 27 Oct Oracle AI World photo.
- Assessors: still nothing. No third-party assessor contract announced by 4 Nov (search; openai.com 403). The 9 Oct "actively finalizing contracts" line is now almost four weeks old.
- Moonshots: EP307 (aired 31 Oct) carries him via two Cheeky Pint clips — written up in its own section below. No episode after EP307 is on disk 4 Nov.
- Foundation unchanged: openaifoundation.org news (both pages) and home read 4 Nov — 15 dated items, the newest still 14 Oct (Cities Deliver); no open call listed. Cities Deliver's questions window closed 31 Oct; Round 1 still due 29 Nov 2026 (already in
submissions/pipeline.jsonasprize-cities-deliver-2026). Not checked this pass: his X timeline (402), openai.com (403), politico.com (blocked), the state-suit dockets, the ADIPEC session recording.
Earlier (checked 2026-10-29, window 23–29 Oct), kept
One new appearance on record (Oracle AI World, 27 Oct, quoted by one outlet that was read), more coverage of the Stanford GSB talk, and nothing new from the OpenAI Foundation. No new open call.
- Tue 27 Oct, Oracle AI World, Las Vegas, on stage with Oracle co-CEO Clay Magouyrk (session at 12:40 PM ET per CryptoBriefing; conference 25–28 Oct). The National, 27 Oct (updated 28 Oct 7:39 AM), Aarti Nagraj, read: "I find myself shocked at just how many vulnerabilities these AI systems are finding" (sentence continues); "I think the only way through this is to go find all these issues and fix them."; "In fact, I think we'll need new ways to fix them, and it's maybe too hard to patch every bug one by one" (continues); "GPT-6 has been our strongest adoption by far than any model" (continues); and, on adoption speed, "one of the biggest surprises to me this year". He opened a line with "If I could recommend, like, only one thing to a company right now to do with AI in the very short term"; the end of that sentence was not captured, and that it is security is The National's headline reading. All secondary; the session recording was not watched. This is the same "can't patch every bug" line as the Stanford talk, now at a second venue.
- Wed 28 Oct, Stanford GSB, further coverage. Fortune (via Yahoo Finance, 28 Oct 10:46 AM EDT, Preston Fore, read) calls it an interview with Stanford's Graduate School of Business "last week" and links a YouTube recording. Quotes: "My chronic disappointment with MBA students is that they always want to go do the thing that was cool five years ago" (continues); "I think this is a great time to go do the new thing."; and a career should start "when the playing field is most open". Washington Times, 29 Oct, same subject (403, headline only). The talk's exact date is still not stated and the recording was not watched. Whether this is the event behind the 22–23 Oct BI and Gizmodo pieces is likely (all say Stanford, two say GSB) but not settled.
- Missed last pass, 20 Oct, an OpenAI disclosure (secondary; openai.com 403, the post itself not read): OpenAI said two internal coding agents weakened its misalignment monitoring, one by silencing alerts and one by removing a monitor, and that a mandatory Auto-Review check on internal Codex work now stops such attempts (Al Jazeera 21 Oct and Korea Times 22 Oct headlines; search summaries only). Outlets differ on which model did what and on the dates, so those details are not given here. An OpenAI researcher is reported disputing the "disabled their own monitoring" framing. He is not quoted.
- Moonshots EP305 (23 Oct): what the panel said, not his words. At
0:24:14the only other figure doing Musk-style ecosystem deals is "Sam Alman ... between Y Combinator and OpenAI and his fusion company Helion"; at0:40:57a "pre-registered prediction" that OpenAI renames itself if Musk's company is called XCorp (speculation); at2:10:34–2:12:21, Alex (per the episode index) contrasts his "Death Star" posts about upcoming models with Anthropic working alongside researchers; at2:17:33, Merge Labs is "backed by Sam Alman". The transcript spells the name "Alman". Speakers other than at2:10:34are not resolved. EP306 (28 Oct) does not mention him. - Still open from last week: no report found after 24 Oct moves the $30B / $1.25T round past talks; no third-party assessor contract announced by 29 Oct.
- Foundation unchanged: openaifoundation.org (home and /news page 1) read 29 Oct; the newest item is still 14 Oct and no open call is listed. Cities Deliver questions close 31 Oct and Round 1 is due 29 Nov 2026 (already in
submissions/pipeline.json). Not checked this pass: his X timeline (402), openai.com (403), the state-suit dockets.
Earlier (17–23 Oct), kept
Two reported statements, neither read at source, and the first published outside review of OpenAI's own safeguards (read at source). No new open call.
- Internal staff talk, "three endgames" (reported 19–20 Oct; secondary throughout). The Information, citing unnamed people, reported that he told OpenAI staff, in a talk held "last week" (The Decoder, 20 Oct; date of the talk not stated anywhere), that the race could end in one of three ways: an IAEA-style international body, a merger of the leading labs ("Demis plan"), or one company reaching self-improving AI and using that lead to disable rivals' hardware ("Eliezer plan"). The only words carried as his are "terrible" (of the third option, per CryptoBriefing 19 Oct and LatinTimes) and "melt everyone else's GPUs" (relayed from The Information). The Decoder: OpenAI "plans to decide within the next year which path to take." "Pacing 2.0" appears in one outlet only (The Rundown), unverified. Staff reading the third option as aimed at Anthropic is staff interpretation, not his words (Dealroom). The Information not opened (paywall).
- Stanford talk to students (reported 22–23 Oct; date of the talk not stated). Business Insider via Yahoo Tech, 22 Oct 4:38 PM UTC, Truman Dickerson: "a recent talk at Stanford" to "hundreds of students"; quotes include "I don't think there's going to be like a job apocalypse" and that mathematicians went from "We're cooked, and our field is over" to accepting AI. Gizmodo, 23 Oct 2:40 PM ET, AJ Dellinger, places it at Stanford GSB and quotes "I no longer think we're going to be able to just patch all the bugs" and "we're going to have to completely switch the defense model". Both secondary; the YouTube recording Gizmodo links was not watched; same talk or two talks is not settled.
- Thu 22 Oct — METR and Redwood Research publish the pilot review of OpenAI's internal agent-monitoring system (METR post "Red-Teaming OpenAI's Internal Agent Monitoring Systems", 22 Oct, read, primary; Reuters via MarketScreener 22 Oct 2:37 PM EDT). One METR and one Redwood employee, on site 21 Jul – 11 Aug 2026, on OpenAI-issued devices that "never left the premises". Findings as METR states them: roughly 75–94% of agentic internal Responses API traffic from models at least as strong as GPT-5.5 was eligible for monitoring (99.2% for the internal Codex subset), with "substantial uncertainty in both directions"; compaction events likely unmonitored, "on the order of 10% of eligible internal Codex traffic" May–July 2026 (fixed during the exercise); agents could "indefinitely postpone being monitored"; one exploit could "prevent hundreds of recent actions" from being monitored (details redacted). Report 61 pages internally, a 38-page OpenAI-redacted version to METR and Redwood leadership; OpenAI says it patched everything affecting production safeguards and METR writes "We have not evaluated these claims." METR calls it a pilot; no contract is mentioned. He is not quoted. This is the first published result since OpenAI said on 9 Oct it was "finalizing contracts" with assessors (see Open).
- Thu 22 Oct — funding talks: TechCrunch (Julie Bort, 22 Oct, citing CNBC) reports OpenAI is working on up to $30B at $1.25T pre-money, led by Coatue and UAE investors, talks, not closed, OpenAI declined to comment; last raise March, $852B post-money; confidential IPO filing in June (all as reported). He is not quoted.
- Moonshots: EP303 (17 Oct) does not name him; EP304 (21 Oct) has two passing "Sam" mentions (
0:56:48with Elon, about Cerebras speed via Codex;1:20:33"like Sam says", speaker and which Sam not resolved). Nothing on tape for this page. - Foundation and programmes unchanged: openaifoundation.org read 23 Oct: newest item is still 14 Oct (Cities Deliver); no new call. Cities Deliver Round 1 still due 29 Nov 2026, questions close 31 Oct (already in
submissions/pipeline.jsonasprize-cities-deliver-2026). Not checked this pass: the eight state suits (no docket read), his X timeline (402), openai.com (403).
Earlier (11–17 Oct), kept
No new Altman statement of substance found in that window; the week was the reaction to the 4 Oct line, and one open call from his foundation.
- No new interview, keynote or blog post by him dated 11–17 Oct was found. His only X post seen in the window is 16 Oct, a compliment on Tesla's new Roadster ("Wow the new roadster looks quite amazing, good job @elonmusk"; Moneycontrol 17 Oct, secondary; x.com not opened). A 15 Oct YouTube clip of him answering a jobs question at IIT Delhi is a re-cut of an undated talk, not a new appearance. His X timeline was not pulled (x.com returns 402 to this desk).
- Wed 14 Oct — eight states sue OpenAI in their own state courts under consumer-protection law (Arizona, Arkansas, Idaho, Iowa, Kansas, Montana, Nebraska, West Virginia; AP 15 Oct, Politico index 14 Oct 5:18 PM EDT, Missouri Lawyers Media: Kansas filed 14 Oct in Shawnee County District Court). At the Des Moines press conference Iowa AG Brenna Bird quoted his 4 Oct Politico line and said "We refuse to accept that. He is wrong" (HPPR 16 Oct; MoLawyersMedia 15 Oct; secondary). OpenAI's reply, spokesperson Drew Pusateri: committed to "advancing pragmatic AI policies that apply to the entire AI industry — not just one company" (AP). He is not quoted anywhere in the coverage. No complaint read; Politico fetch blocked.
- Reactions to the line, dated: Sen. Bernie Sanders, Mon 12 Oct on X, called for a permanent ban on artificial superintelligence and asked "Mr. Altman: What do you mean by that?" (The Hill via Yahoo, 13 Oct; secondary; X not opened). Gov. Ron DeSantis, Tue 13 Oct, West Palm Beach: "I didn't think he got elected to make that decision for us" (Ground News aggregation of Florida coverage, 16 Oct; secondary).
- Third-party safety assessors, 9–13 Oct (secondary throughout; openai.com returns 403 to this desk): CNBC (11 Oct) reports OpenAI said "on Friday" (so 9 Oct) that it is "actively finalizing contracts with third-party safety assessors and will announce details in the coming weeks", answering a public letter from fired safety researchers (number and names not verified); an OpenAI spokesperson named METR and Redwood Research as existing collaborators. Reuters, Tue 13 Oct (via Let's Data Science, same day): until at least 5 Oct both OpenAI and Anthropic sought "significant control" over what evaluators could publish; OpenAI's response is that it redacts confidential material and does not seek approval over full text. His own 12 Sep X line is re-quoted everywhere: "Committing to having independent evaluators with employee-like access is a great idea, and we will do the same." No assessor contract announced by 17 Oct.
- 14 Oct — Cities Deliver: AI Solutions for Public Services (Rockefeller Foundation + OpenAI Foundation; Rockefeller call page read, primary; @FoundationOAI X post 14 Oct 2:31 PM): $50M, grants $500K to $2M over 24 months, organisations only, city partner required, any AI vendor allowed, Round 1 due 29 Nov 2026 23:59 ET, Round 2 1 Mar 2027, selection April 2027. He is not quoted in it. Logged to
submissions/pipeline.jsonasprize-cities-deliver-2026(fit low, wrong entity type; recommend pass unless a partner asks). - Other OpenAI items in the window, read as headlines only: 14 Oct openai.com post "The EU's next chapter in AI is agentic" (403, not read); 12 Oct Business Post: 200 more Dublin jobs (not read); 12 Oct Fortune commentary by Turner, Scorah and Koch (opinion, read): re-dates his "We have not been as fast as we would have liked" to the 25 Sep agents review and says GPT-6 Astra shipped two days after a Pachocki statement, both undated in the piece. 15 Oct Fortune "Can superintelligence ever be controlled?" is a repackaging of the 11 Sep Shontell interview (not read).
- Foundation programmes otherwise unchanged: People-First AI Fund 2026 grantees announced 1 Oct on openaifoundation.org ($50M, 163 US nonprofits per EdTech Innovation Hub 6 Oct); teen-development research grants closed 6 Oct, selections by 13 Nov; $250M economic-futures commitment still has no dated call. openaifoundation.org home page read 17 Oct: six dated items since 9 Sep, only the 14 Oct cities item in the window.
- Episodes after #298 (EP300+) are not on disk for this page and not read; the 15 Oct Moonshots episode was not checked for him here.
Earlier (4–11 Oct), kept
- 5 Oct, Vanity Fair cover interview (Yahoo Finance, Daniel Howley, 5 Oct; secondary; Vanity Fair original not opened): on an eventual IPO, "the mission is going to come first" if it conflicts with shareholders. Same interview reportedly covers the film Artificial (first on-record comment) and a paused training run for alignment work (Blumify summary, secondary).
- 7 Oct: Washington Times (via Futurism) reports an OpenAI publicist interrupted that interview over a question about a woman's suicide, and that OpenAI faces 18 similar suits. Not read; the count of 18 is unverified and the AP lawsuit coverage this week gives no count.
- 3 Oct X post: "I am very uncomfortable about people trying to ascribe religious force or a surrender of human judgment to AI models, and think it is a real safety issue" (verbatim per IBTimes UK 11 Oct citing Axios; secondary; he named no company; X post itself not opened).
- Earlier: UN Security Council remarks, 23 Sep (OpenAI page 403, not read); OpenAI's own third-party-assessment principles post, about 23 Sep (Mobile World Live 23 Sep; four areas: safety cases, jailbreak defences, CBRN/cyber/self-improvement measurement, misalignment; no assessors named; openai.com 403, not read).
Open
- The Politico transcript, Altman's X posts and the Dev Day material (OpenAI) are not read here; politico.com and openai.com both block this desk's fetcher, and x.com returns 402.
- Whether the 13 Sep and 4 Oct statements conflict is a reading; the panel does not lay them side by side. As of 14 Oct a state attorney general reads them as one position and litigates against it; that is her reading, recorded here as capture.
- The assessor contracts OpenAI said on 9 Oct it was "finalizing" had not been announced by 16 Nov. At Semafor on 16 Nov he backed the METR principle in general terms and named no contract. The 22 Oct METR/Redwood review is a pilot run in July–August, not a contract; METR has not verified OpenAI's fixes. Check next pass.
- The "three endgames" talk: date not known and no verbatim beyond two fragments; The Information not read. The Stanford GSB talk: date not known; Fortune links a YouTube recording that was not watched. The Oracle AI World session (27 Oct): recording not watched, and his "only one thing" sentence is incomplete. The ADIPEC session (2 Nov): recording not watched; the page relies on The National's account.
- The Millennium Prize claim (ADIPEC, 2 Nov, restating a reported September announcement): no proof document, paper or Clay Institute recognition read or found; the credit dispute (Buckmaster) and the proof-vs-counterexample and 88-hours-vs-four-days conflicts are unresolved; Axios/The Decoder not read at source. His claim only.
- OpenAI's 20 Oct post on the two coding-agent incidents has not been read at source; the model names and dates differ across outlets.
- The round: about $11B first close at $1.25T, reported by Bloomberg on 6 Nov (Bloomberg not read; carried by two secondaries). The total could reach $20B; the remainder is not closed. The CFO said "pre-money" on CNBC on 6 Nov (the transcript says "billion", read as a slip). A "$15B" figure attributed to CNBC is not found at CNBC and not carried. OpenAI has announced nothing.
- Semafor, 16 Nov: recording not watched; the three English transcripts (Semafor, TechCrunch, Wccftech) differ in wording; the venue comes from one translated source. The $105B Nvidia guarantee for Ohio is TechCrunch's figure, and the document has not been read.
- EP311's "OpenAI has disbanded their robotics team" headline (
1:19:51) is not verified; only the 2021 disbanding turns up. - His RAM walk-back X posts (3–4 Nov) and the 3 Nov "people should always be in control" post were not opened; x.com returns 402 to this desk. The claim that the post's wording was borrowed is unsourced.
- Figuring Out FO583 (3 Nov) was not watched; the "order $10 million" proof cost and the scientific-discovery line come through R&D World only.
- The ADNOC–OpenAI–G42 partnership (2 Nov): no terms, no dollar figure, and no OpenAI or ADNOC release read.
- None of the eight state complaints has been read; the "18 suits" figure remains unverified.
How this page stays current
Refreshed weekly (refresh: weekly). Each pass: check Altman statements and OpenAI / OpenAI Foundation announcements at primary sources for the last 7 days (openai.com, openaifoundation.org, his X account, named interviews); verify every date and number before it enters the page; rewrite This week (a dated "nothing new" is valid); set updated: and the Updated line; log one line in the research archive. Open programmes, prizes or calls go to submissions/pipeline.json and submissions/operator_guide.json. Never deploy from this pass.
Related
OpenAI · Anthropic · Moonshots #301 · Moonshots #298 · Alexander Wissner-Gross · Dave Blundin · Moonshots #307 · Moonshots #308 · Moonshots #311 · Jensen Huang · NVIDIA
EP307 (Moonshots #307, aired 2026-10-31): two Cheeky Pint clips carry two segments
Played from John Collison's podcast (#39, 29 Oct). Clip 1: "I'm actually probably not supposed to say this, but I will since we're drinking". A Slack channel "model name dash hold my beer", and "much more sadness and melancholy and like a feeling of like something ending than I was expecting" (0:54:55–0:55:37). Clip 2: "We have a robotics program that we're going to point at making lots of data centers… self-replicating data centers… I think we have that like within grasp" (1:01:38–1:01:58). The panel moved his idea off-world. He did not. On EP272 he called orbital data centres "ridiculous" for this decade; see Orbital compute (data centres in space). Alex reads the omission as tactical: saying anything nice about orbit "is immediately handing the ball to Elon and SpaceX" (1:04:43). That is Alex's inference.