‘Sam Altman is an historic con artist’ | Ed Zitron

This video argues that OpenAI and Sam Altman are running "the largest con within AI," driven by deceptive hype, meaningless safety rhetoric, and unsustainable economics that the media refuses to scrutinize. It highlights how Altman, exposed as a liar and fantasist by a New Yorker piece, exploits investors' greed while media coverage mystically frames AI rather than addressing real harms like misinformation and environmental damage. The analysis also exposes the industry's deliberate misuse of the word "training," revealing it as a permanent, escalating operational cost—projected at over $120 billion over two years—not a one-time research expense, with a massive funding gap hidden by non-GAAP accounting. Both OpenAI and Anthropic are condemned as identical "cons" that treat AI safety as a marketing tool, with CEOs like Altman and Dario Amodei caring only "as much as they have to." Ultimately, the video calls for real financial transparency, arguing an IPO would expose the bubble's "innards," while mainstream coverage dodges the hard economics that threaten powerful market interests.

• ▶ 0:00 OpenAI's financial situation is described as "the largest con within AI"—investors won't buy tens of billions in bonds for a company that only needs money to lose more money.
• ▶ 0:43 The New Yorker exposé, based on over 100 interviews, mainstreamed the view that Sam Altman is a deceptive "incompetent sociopath" who is actually a fantasist, not a genius.
• ▶ 2:08 Altman is just "a node"—a guy skilled at exploiting powerful people's greed (Microsoft, SoftBank, Oracle, etc.)—and the only real known fact about him is that he is a liar.
• ▶ 3:55 Media coverage still frames AI in mystical terms (e.g., "Can we trust Sam Altman with powerful AI?"), ignoring that the real dangers are specific harms like misinformation, theft, and environmental damage, not an uncontrollable superintelligence.
• ▶ 4:44 The AI scare tactic is a marketing strategy: media becomes a co-conspirator in OpenAI's con by repeating grandiose claims about AI agents and coding tools that don't actually work as described.
• ▶ 7:21 The entire AI bubble depends on obfuscating what AI actually does, costs, and can achieve—and the media is complicit in repeating hype that fuels a self-reinforcing cycle of investment and unfounded headlines.
• ▶ 8:07 Altman has “got away with it” largely because of broader OpenAI problems and because he is “a genuinely unlikable guy”; no coordinated smear campaign is needed since “Altman does it himself.”
• ▶ 8:49 The exposé isn’t gaining traction because it avoids “the economics and the efficacy” of AI, which these personality-focused articles never address.
• ▶ 9:16 Digging into real economics is tough and threatens powerful market interests, while attacking Altman personally is narratively satisfying and faces far fewer vested interests trying to stop it.
• [10:06–10:27] The piece dodged the real problem: AI is sold on theoretical capabilities, not actual results, and the industry's economics are built on those promises — signs of a bubble no one wants to confront.
• [10:46–11:21] Anthropic is "OpenAI in a different hat": Dario Amodei is just as bad as Altman, and Anthropic mirrors OpenAI's deals, IPO rush, customer treatment, and annoying CEO.
• [13:04–14:00] AI safety is used as a smokescreen across the industry — Anthropic, Meta, Microsoft, Oracle, and Google all care about safety only "as much as they have to."
• ▶ 13:56 OpenAI and Sam Altman treat AI safety as a marketing tool, not a genuine priority—they care about safety only “as much as they have to.”
• ▶ 14:41 Real safety is impossible when hallucination-prone AI is deployed in high-risk fields, and OpenAI’s own research admits hallucinations can’t be removed without a “new kind of mathematics.”
• ▶ 15:20 Altman’s safety talk is a lie; he really cares about money, power, and access, making him just another tech mogul like Bill Gates—except Gates actually made things.
• ▶ 15:42 Zitron argues Sam Altman is not fundamentally different from other Big Tech CEOs, except that Bill Gates actually made stuff; Altman resembles management-consultant types like Nadella, Pichai, and Cook.
• ▶ 16:00 Altman’s “safety” rhetoric is hypocritical: at the start of the Iran war he was enthusiastically seeking Department of Defense contracts and classified information.
• ▶ 16:18 Anthropic is no different, having worked with the Department of Defense since June 2024; all AI CEOs just do “different impressions of a safety-focused guy.”
• ▶ 16:04 OpenAI CFO Sarah Friar privately said the company is not ready to IPO in 2026 and unsure if growth can cover its compute spending—a major red flag.
• ▶ 16:26 Zitron argues Altman is rushing toward an IPO mainly to provide "exit liquidity for everyone involved in the con," not because OpenAI is financially sound.
• ▶ 18:30 Zitron calls the IPO push a "terrible idea," noting OpenAI is caught between investor pressure to go public and the company's own unpreparedness, while huge annual funding needs and junk-level credit prospects make the math unsustainable.
• ▶ 19:49 R&D should be counted as a running cost, and calling it “training” is “the largest con within AI.”
• ▶ 19:59 “Training” is an intentional mislabel that frames ongoing compute expenses as capital expenditures, when it actually covers bug fixes, model drift, and new model development.
• ▶ 20:18 Training never stops — it is a permanent, escalating operational burden, not a finite one-time R&D project.
• ▶ 20:33 OpenAI is projected to spend over $120 billion on training in the next two years, per a Wall Street Journal report.
• ▶ 20:44 A funding gap exists: OpenAI raised $122 billion, but only about $60 billion is believed to be real cash—far short of the projected spend.
• ▶ 20:51 Key unanswered questions remain over how OpenAI will fund the $120 billion, with the host criticizing the WSJ for not pressing these obvious points.
• ▶ 21:02 The word “training” is itself a con: it implies a temporary prep phase, but it’s actually an ongoing, indefinite expense.
• ▶ 21:14 Training isn’t CAPEX, but it has the same problem—spending keeps rising and no one can answer “When do they stop exactly?”
• ▶ 21:41 Training effectively never stops: OpenAI and Anthropic spend tens of billions every year, and their margins get worse annually, not better.
• ▶ 21:50 OpenAI and Anthropic's margins have worsened every year, which is problematic given their massive valuations.
• ▶ 21:58 Both companies rely on non-GAAP metrics and "wacky margins" rather than standard audited financials, reducing transparency.
• ▶ 22:06 Zitron wants them to file for IPOs so their internal finances face public scrutiny — "Show me the innards. Show me the guts."
• ▶ 22:11 The host challenges Anthropic's framing of Claude as "too dangerous to release," noting the contradiction between safety claims and business incentives.
• ▶ 22:17 The conversation questions whether the lockdown is genuine safety concern or a PR stunt to hide massive compute costs, referencing steep spending ratios like $25 to $1.
• ▶ 22:34 Ed Zitron bluntly concludes it's all marketing, dismissing the safety narrative as a way to build mystique while obscuring real operational costs.
• ▶ 22:32 The entire "Claude Mythos escaped the sandbox" narrative is dismissed as marketing, with even the official system card framed as part of that PR exercise.
• ▶ 22:37 The "escape" claim is ridiculed as absurd, since the supposed breakthrough was only noticed when the model "emailed me eating a sandwich."
• ▶ 22:51 In reality, Claude Mythos did not escape; it was run in a separate environment, given computer access, and explicitly told to break something and email the result—so the story was a controlled test repackaged as a dramatic event.
• ▶ 23:08 The speaker argues that the product is being treated as if it is "alive," fueling an irrational, hype-driven mystique around it.
• ▶ 23:12 They single out the deliberate narrative of releasing only to 40 companies, mocking it as artificial scarcity.
• ▶ 23:15 The tone is sarcastic: presenting this tiny rollout as "reserved" and "careful" is really a contrived PR move to manufacture exclusivity and prestige.
• ▶ 23:18 Anthropic leaked the source code of Claude Code, contradicting their claim of having a "bug-finding machine" so good it "couldn't find bugs."
• ▶ 23:30 The speaker dismisses the defense that the tool wasn't used on Claude Code, asking "Why?" and noting that cheaper models have already found similar bugs in the leaked code.
• ▶ 23:43 The headline "27-year-old bug" is dismissed as mundane and expected in open source, not proof of extraordinary AI capability.
• ▶ 23:56 Mythos only found bugs—it did not fix them; humans still had to do all repair work.
• ▶ 24:04 LLM-based bug reports have a serious false-positive problem: typically only one-third of findings are genuinely useful, with two-thirds being noise.
• ▶ 24:21 The high false-positive rate makes the tool less attractive, especially with a high price tag—the speaker starts to note "Except this one's $125..." before cutting off.
• [24:25–24:52] The banking meeting over the Claude Mythos story is suspicious: JP Morgan, a company with actual access to Claude Mythos, was excluded from the briefing, even though it could confirm or deny the alleged danger.
• [24:52–25:08] Partner reactions contradict the fear narrative—if the model were truly terrifying, companies like JP Morgan, CrowdStrike, and Microsoft would be publicly and internally freaking out, but they are not.
• [25:08–25:29] Insider contacts at Microsoft, CrowdStrike, Oracle, and elsewhere are silent, with only muted reactions; the story lacks key technical details and fails to compare Claude Mythos to Claude Opus with the same harnesses, undermining its claimed uniqueness.
• ▶ 25:31 Ed Zitron suggests Anthropic is building hype toward its IPO, calling the situation "very muddy."
• ▶ 25:48 Zitron calls out Amodei's unsubstantiated claim that 50% of white-collar jobs have been replaced in 12 months, repeated at Davos.
• ▶ 26:01 Zitron dismisses the rhetoric as "anyone can make stuff up" and expresses fatigue with such predictions.
• ▶ 26:11 Zitron argues the entire basis of "Methos" fear-mongering is simply that "we can't use it," making inaccessibility the sole premise for calling it scary.
• ▶ 26:22 He dismisses the coverage as "key jingling"—a distraction tactic mocking the idea that restricting access automatically equates to power or danger.
• ▶ 26:31 Zitron pivots to OpenAI's rumored "Spud" model, mocking the $200M TBP deal and the terrible codename, noting no engineer would feel inspired working on "spud," whereas "Mythos" at least sounds fun by comparison.

• ▶ 27:06 Zitron argues LLMs are as much a marketing, psychological, and religious concept as a technology—the vagueness is “part of the con” and genuinely unsafe.

• ▶ 27:29 Sam Altman is credited with exploiting how people extrapolate meaning from LLMs, while many powerful figures never even use them—making AI “the perfect con.”

• ▶ 27:47 Anthropic is “becoming OpenAI 2,” and OpenAI is “WeWork 2,” meaning Anthropic is effectively “WeWork 3”—the next hype-driven bubble.

• ▶ 27:54 Anthropic's "human evaluation" claim is vague and contradictory: they cite human involvement but refuse to explain it, while simultaneously touting the model as powerful and scary.
• ▶ 28:22 The security rationale for not releasing is weak, since open-weight models would likely replicate the same capabilities within months anyway.
• ▶ 28:48 The real reason for withholding the model is probably to avoid immediate "BS testing" that would expose inflated capability claims.

• ▶ 29:19 OpenAI claims 40% of its business is enterprise, projected to become majority, but the speaker is skeptical: "OpenAI says a lot of crap, so who knows?" — Operation Glass Wing is just an enterprise push, not a mission shift.

• ▶ 29:28 The launch sparked a "vacuous" Twitter discourse framing it as withholding models from regular people ("models only for the rich"), which the speaker dismisses as overwrought and absurd — calling the reaction "pathetic" at ▶ 29:47.

• ▶ 29:50 The speaker uses a Salesforce analogy: imagining Mark Benioff theatrically claiming his Einstein AI threatened him with a gun — highlighting how silly the "spooky enterprise launch" framing is for a routine product rollout.

• ▶ 30:06 Zitron hopes the company goes public because it will be much more difficult to keep up the hype and narrative manipulation.
• ▶ 30:13 Each major announcement or stunt could temporarily send the stock running higher.
• ▶ 30:16 But the company's underlying economics will ultimately send the stock "to hell."
• ▶ 30:23 Ed Zitron asks viewers who enjoyed the episode to like and subscribe for more of The Tech Report.
• ▶ 30:29 He reminds the audience that episodes are also available as a podcast wherever they get their podcasts.

• ▶ 0:00 The Largest Con and the New Yorker Exposé (0:00 - 3:13) - - Ed Zitron introduces the biggest AI con and assesses the New Yorker's Sam Altman exposé.
• ▶ 3:13 Media Complicity and the AI Bubble's Obfuscation (3:13 - 7:52) - - The hosts argue media scare tactics prop up OpenAI while revenue claims and technical explanations don't hold up.
• ▶ 7:52 Altman's Damage and Why the Story Stalls (7:52 - 9:36) - - They question whether the exposé hurts Altman and explain why it's not gaining traction.
• ▶ 9:36 The Piece's Gaps and Anthropic's Safety Hypocrisy (9:36 - 14:00) - - The piece is important but dodges bubble economics; Dario Amodei and Anthropic repeat OpenAI's safety doubletalk.
• ▶ 14:00 AI Safety as Marketing and Mogul Hypocrisy (14:00 - 15:46) - - Safety talk is often marketing, hallucinations make risky deployments questionable, and Altman is just another mogul.
• ▶ 15:46 AI CEOs, OpenAI's Finances, and Training-Cost Con (15:46 - 22:11) - - Criticism extends to AI CEOs' defense work, OpenAI's cash burn/IPO push, and R&D/training cost framing.
• ▶ 22:11 Claude Mythos and Anthropic/OpenAI IPO Hype (22:11 - 30:34) - - The segment unpacks Claude Mythos marketing, bug-report issues, and the hype around Anthropic's IPO and OpenAI's Spud/TBP.