Jim Cramer spent Tuesday night talking up a landmark artificial intelligence IPO. By Wednesday morning, he was comparing that same company’s safety math to a surgery no rational patient would choose. Fewer than 14 hours separated the two takes, and the whiplash traces back to a single number: a greater than 10% chance, in the words of an Anthropic safety researcher, that AI kills people.
The episode, first reported by 24/7 Wall St. on September 9, 2026, captures a tension that has been building all year in AI circles. The same lab racing toward a reported $2 trillion IPO valuation employs researchers who are willing to put double-digit odds on their own technology causing mass harm. Cramer, a CNBC host not known for hedging his opinions, tried to hold both ideas at once on live television and visibly struggled.
Cramer’s Two Takes on Anthropic, 14 Hours Apart
On Tuesday night’s Mad Money, Cramer told viewers that Anthropic, still a private company, could be a gigantic IPO, one big enough that investors would sell other holdings just to raise cash to participate. That framing lines up with months of Cramer commentary treating Anthropic as the AI trade’s next major public listing.
By Wednesday morning on Squawk on the Street, the tone had flipped. Cramer was reacting to a number circulating from inside Anthropic itself: a safety researcher’s public estimate that AI carries more than a 10% chance of killing people. He posted a shorter version of his reaction on X at 9:42 a.m., framing it bluntly as a 10% chance everyone dies from AI and stating he did not like those odds. The full CNBC segment ran with a comparison to elective surgery, which we break down next.
The Surgery Analogy That Cramer Used to Reject 10% Odds
Cramer’s on-air argument, delivered during the September 9 Squawk on the Street broadcast, leaned on a medical comparison. He argued that if a surgery carried more than a 5% fatality rate, no reasonable patient would agree to it. A 10% figure, he noted, is double that threshold, which made the odds unacceptable to him personally. He also said he was surprised the researcher behind the estimate still works at Anthropic, referring to CEO Dario Amodei by his first name.
Jim Cramer had already staked out a bullish position on Anthropic’s valuation before the safety comments surfaced. Reacting to reports that Anthropic could command a $2 trillion valuation, Cramer wrote on air, “We will hear all day today that if Anthropic can command $2 trillion than everything is out of hand. It’s not fraught, though, if they have the revs” (Yahoo Finance). That defense of the valuation, paired with his surgery-odds rejection a day later, is what makes the reversal notable. He is not walking back the price tag. He is walking back his comfort with the company behind it.
Who Is Evan Hubinger, and What Did He Actually Say
The researcher at the center of the story is Evan Hubinger, described across multiple outlets as heading alignment science or alignment stress-testing work at Anthropic. Hubinger posted publicly on X, replying to a colleague’s comment about AI risk. His full post read: “Jacob is correct here—we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade” (X / Evan Hubinger).
Hubinger did not stop at the raw probability. In the same thread, he added a second, arguably more uncomfortable point for a company selling itself as the safety-focused AI lab: “I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to” (X / Evan Hubinger). That is a sitting Anthropic employee stating, on the record, that his own employer lacks a working plan for the exact problem it was founded to solve.
A Second Researcher’s Exit Adds to the Pressure
Hubinger’s “Jacob is correct here” line referenced a separate Anthropic researcher, identified in 24/7 Wall St.’s reporting as Jacob Coxon. According to that report, Coxon had resigned from Anthropic and wrote that Anthropic and OpenAI are racing toward self-improving superintelligence in a way that gambles with humanity’s safety. Two researchers going public with similar concerns inside a matter of days is a different signal than a single outlier post, and it is part of why the story crossed over from AI-safety circles into financial media within hours.
Anthropic’s IPO Timeline: What’s Confirmed and What’s Rumored
Separating fact from speculation matters here, because Cramer’s Tuesday-night comments were about a deal that does not officially exist yet. Anthropic remains a private company. It has not filed a public S-1 with the SEC, and multiple reports describe only a confidential draft registration submitted earlier this year, which is a standard early step that does not commit a company to a public listing or a timeline.
Reported Valuation Estimates
Reported figures for a potential Anthropic listing have ranged widely depending on the outlet and the week. Some reports place pre-IPO investor discussions around $965 billion, while later reporting has floated a target as high as $2 trillion for a late-September or early-October window. None of these numbers come from Anthropic itself, and all of them should be read as reported estimates rather than confirmed pricing.
What Anthropic Hasn’t Said
As of September 9, 2026, Anthropic has not confirmed an exchange, a ticker, a share price, or a final listing date. The company has previously told reporters, in earlier coverage from around December 2025, that it had not decided when or even if it would go public. That caveat still applies. Nothing about the Hubinger post changes the legal status of the IPO, but it does change the narrative Cramer and other commentators are building around it.
| Reported detail | Figure or status | Confirmed by Anthropic? |
|---|---|---|
| Confidential draft S-1 | Filed with the SEC earlier in 2026 | Reported, not confirmed by company |
| Public S-1 filing | Not filed as of Sept. 9, 2026 | Confirmed absent |
| Reported investor-meeting valuation | ~$965 billion | Reported only |
| Reported target valuation (high end) | Up to $2 trillion | Reported only |
| Target listing window | Late Sept.–early Oct. 2026 | Reported only |
| Exchange, ticker, share price | Not announced | Confirmed absent |
Why a 10% Estimate Rattles Wall Street More Than AI Circles
Inside AI safety research, probability estimates for catastrophic outcomes, sometimes shorthanded as “p(doom),” have circulated for years in surveys, blog posts, and internal memos. A double-digit figure from an Anthropic researcher is not new in that world. What is new is a mainstream financial commentator reading that number on live television and applying it to a company he had just told his audience to treat as an investable, IPO-bound business. Cramer’s surgery framing translates an abstract safety debate into a decision a retail investor can immediately understand: would you take a 10% chance of dying to get the upside. Stated that way, the number does a lot more work than it would buried in a research paper.
Dario Amodei’s Own Numbers on AI Risk
Hubinger is not the first person at Anthropic to put a number on catastrophic risk. In comments reported in June 2026, CEO Dario Amodei described roughly a one-in-four chance that AI development goes badly, against a three-in-four chance that it goes well. That is a lower probability than Hubinger’s more than 10% figure, but it is the same basic exercise: a senior Anthropic figure assigning odds to an outcome most companies would never discuss in public, let alone quantify. Cramer’s surprise that Hubinger still works at the company, given his estimate, reads differently once you know the CEO has been running the same kind of math publicly for months.
Market Reaction: Meta and Airbnb as the Public AI Trade
Because Anthropic itself has no public shares to trade, investors reacting to the story on September 9 had to route that reaction through public AI proxies instead. Two names moved in opposite directions the same morning, and both are useful for showing how little the safety headlines actually affected broader AI sentiment.
Meta and Airbnb Move in Opposite Directions
Meta Platforms posted Q2 2026 revenue of $60.80 billion, up 27.96% year over year, though diluted EPS of $6.18 missed consensus by roughly 14%. Capital expenditures hit $30.116 billion for the quarter, with full-year capex guidance now sitting at $130 billion to $145 billion, and free cash flow compressed to $784 million. On September 9, Meta traded up 5.7% intraday and was up 12.1% for the week, driven by its Muse AI agent launch, a rally that shrugged off the Anthropic safety story entirely.
Airbnb told a different story. Q2 2026 revenue reached $3.608 billion, up 16.54% year over year, with GAAP EPS of $1.27. The company says its AI assistant now resolves roughly 45% of customer support issues without a human agent, cutting support cost per booking by about 16% year over year. Despite those numbers, Airbnb shares traded down 2.93% on September 9 and were off 7.19% for the week, even while still up nearly 25% year to date.
| Metric (Q2 2026) | Meta (META) | Airbnb (ABNB) |
|---|---|---|
| Revenue | $60.80B (+27.96% YoY) | $3.608B (+16.54% YoY) |
| EPS | $6.18 diluted (missed est. by ~14.4%) | $1.27 GAAP |
| AI capex / usage note | $30.116B quarterly capex; FY guide $130B–$145B | AI assistant resolves ~45% of support tickets |
| Free cash flow | $784M (compressed) | Not disclosed in this comparison |
| Sept. 9 share move | +5.7% intraday, +12.1% weekly | -2.93% session, -7.19% weekly |
| Year-to-date | Not disclosed in this comparison | +24.83% YTD |
How Other AI Labs Handle Public Risk Estimates
Anthropic’s willingness to let researchers publish double-digit catastrophic-risk numbers stands out against how most of its competitors handle the same topic. OpenAI, Google DeepMind, and xAI generally address AI safety through official blog posts, model cards, and framework documents rather than individual researchers posting personal probability estimates on social media. That difference is partly cultural. Anthropic was founded by former OpenAI staff specifically around the argument that safety work needed more independence and more public accountability, and Hubinger’s post, whatever damage it does to a single news cycle, is consistent with the safety-first positioning Anthropic maintains on its own newsroom page. The trade-off is exactly what played out this week: transparency that plays well with researchers can read as a red flag to an investor audience unfamiliar with how the p(doom) conversation normally happens inside these labs.
Historical Context: From Private Estimates to Public Statements
Catastrophic-risk probability estimates from AI researchers are not a 2026 invention. Surveys of machine learning researchers going back several years have asked participants to assign odds to existential outcomes, and results have varied enormously depending on wording and audience. What changed this week is the venue. A number that used to live in a survey footnote or a niche forum thread reached a CNBC audience within hours, attached to a specific company on the edge of a trillion-dollar IPO. That is a meaningfully different distribution channel, and it is why a Tuesday-night bull case turned into a Wednesday-morning walk-back so quickly.
Expert Voices on the Anthropic Whiplash
Cramer himself has been the most visible voice defending Anthropic’s valuation even as he distances himself from the safety framing. Earlier in the IPO speculation cycle, he pushed back on the idea that Anthropic waiting for a better market window was itself a warning sign, saying, “This market does not have room at this moment for Anthropic. They should wait until the fall” (24/7 Wall St.). That comment, made months before the Hubinger post, shows Cramer has treated Anthropic’s IPO timing as a market-conditions problem rather than a company-quality problem for most of 2026. The safety estimate is the first input this year that made him question the company itself rather than just the timing of its debut.
The Broadcom Connection: Why This Isn’t Just an Anthropic Story
Cramer has previously tied Broadcom’s bull case directly to Anthropic, framing the chipmaker’s prospects as effectively “Anthropic or bust” in comments around September 3. That framing means a delayed or shrunk Anthropic IPO would not stay contained to one private company’s cap table. It would ripple into the public names investors already hold, particularly suppliers positioned around Anthropic’s compute buildout. A safety-driven pause to IPO plans, even a short one, is the kind of event that could show up in Broadcom’s valuation conversation well before Anthropic ever files a public prospectus.
What Happens Next for Anthropic’s IPO Filing
Three things are worth tracking from here. First, whether Anthropic actually files public IPO paperwork, something it had not done as of September 9. Second, whether additional Anthropic staff go public with their own risk estimates, which would turn Hubinger’s post from an outlier into a pattern. Third, whether the broader AI capex cycle, visible in Meta’s $130 billion to $145 billion full-year guidance, keeps drawing investor interest even as free cash flow at companies like Meta compresses under the spending. All three threads point back to the same underlying question: can a company be simultaneously the most safety-transparent AI lab and a trillion-dollar-plus IPO candidate, or does one of those identities eventually have to give ground to the other.
Five Predictions for the Months Ahead
- Anthropic is unlikely to publicly address Hubinger’s specific 10% figure directly, since doing so would force the company to either endorse or contradict a working researcher’s stated view on its own core product risk.
- Expect more financial commentators to start citing individual Anthropic researcher statements the way they currently cite earnings calls, turning AI safety posts into a new category of market-moving commentary.
- A public S-1 filing, if it comes, will likely arrive with expanded risk-factor language addressing AI safety and alignment more explicitly than prior tech IPOs have needed to.
- Broadcom’s stock narrative will keep getting pulled into Anthropic IPO timing questions given Cramer’s own framing of the two as linked.
- Watch for other labs, particularly OpenAI and Google DeepMind, to face pointed questions from reporters about whether their own researchers hold similar private probability estimates, even if none have published one as directly as Hubinger did.
Frequently Asked Questions
What did Evan Hubinger actually say about AI risk?
Hubinger, an Anthropic researcher, wrote on X that he believes there is a greater than 10% chance of AI killing all humans within the next decade, and that Anthropic does not yet have a working plan to solve alignment for superintelligence.
Why did Jim Cramer change his view on Anthropic so quickly?
Cramer praised Anthropic’s IPO prospects on Tuesday night’s Mad Money, then reacted to Hubinger’s 10% risk estimate the next morning on Squawk on the Street, comparing the odds to an unacceptable surgical risk and saying he did not like them.
Has Anthropic officially filed for an IPO?
No. As of September 9, 2026, Anthropic has not filed a public S-1 with the SEC. Reports describe only a confidential draft filing, and the company has not confirmed an exchange, ticker, share price, or exact listing date.
What valuation is Anthropic reportedly targeting for its IPO?
Reported figures vary widely, from around $965 billion in earlier investor-meeting coverage to as high as $2 trillion in later reports about a late-September or early-October window. None of these numbers have been confirmed by Anthropic.
Who is Jacob Coxon and how does he relate to this story?
According to 24/7 Wall St.’s reporting, Jacob Coxon is a separate Anthropic researcher who resigned and wrote that Anthropic and OpenAI are racing toward self-improving superintelligence in a way that gambles with safety. Hubinger referenced him directly in his own post about AI risk.
What has Dario Amodei said about AI risk odds?
In comments reported in June 2026, Amodei described roughly a 25% chance that AI development goes badly and a 75% chance that it goes well, a separate and lower estimate than Hubinger’s more than 10% figure for a worse-case outcome.
Did the AI risk story affect AI stocks broadly?
Not uniformly. Meta shares rose 5.7% intraday on September 9 on its Muse AI agent launch, while Airbnb shares fell 2.93% the same day, suggesting the safety headlines were not the dominant driver for either move.
Is a 10% catastrophic-risk estimate unusual among AI researchers?
Double-digit probability estimates for catastrophic AI outcomes have circulated in AI safety research and surveys for years. What made Hubinger’s post unusual was its reach into mainstream financial media within hours, tied directly to a company on the edge of a major IPO.
