Case TS-CF4E2B7E6 Sept 2026capabilityCompound claim

AI

“Sam Altman and Dario Amodei both signed statements putting AI extinction risk on par with nuclear war, and have separately given probability estimates of 10-25% for AI-related existential risk, while continuing to build AI systems." (Post caption wording: "Altman and Amodei both signed statements putting AI extinction risk next to…”

Plain restatementBoth Sam Altman and Dario Amodei signed a public statement on AI extinction risk that referenced nuclear war, and each has separately stated a personal probability estimate in the 10 to 25 percent range for existential risk from AI, while both continue to lead companies building frontier AI.

Partially accurate but misleadingConfidence Medium
What this verdict means →

Distortion code this site does not recognise yet: misattribution. Not collectible until the field guide has an entry.

Half of this post checks out and half does not. It is true that Sam Altman and Dario Amodei both signed the Center for AI Safety statement in May 2023, which says reducing extinction risk from AI should be a global priority alongside pandemics and nuclear war. It is also true that both still run companies building advanced AI. But the "ten to twenty five percent by their own admission" line credits one man's number to two. That range is Dario Amodei's, stated in a 2023 interview and repeated at 25 percent at an Axios event in September 2025. Sam Altman has never been recorded giving that range. The only figure ever attributed to him is around 2 percent, second-hand, which he declined to treat as a literal number, and in July 2026 he was asked directly for a percentage and refused to give one, saying the odds are "very low." The post also converts Amodei's phrase about civilization-scale catastrophe into "existential risk," which is a narrower and more dramatic claim than what he actually said.

The drift / as claimed vs as evidenced

Sam Altman and Dario Amodei [drifted from the evidence:] both signed [drifted from the evidence:] statements putting AI extinction risk [drifted from the evidence:] on par with nuclear war, and [drifted from the evidence:] have separately [drifted from the evidence:] given probability [drifted from the evidence:] estimates of 10-25% for [drifted from the evidence:] AI-related existential risk, [drifted from the evidence:] while continuing to build AI [drifted from the evidence:] systems." (Post caption wording: "Altman and Amodei both [drifted from the evidence:] signed statements putting AI extinction risk next to [drifted from the evidence:] nuclear war. Then kept building, [drifted from the evidence:] because stopping hands it to a rival. Ten to twenty five percent by their own admission.")


[added by the neutral restatement:] Both Sam Altman and Dario Amodei signed [added by the neutral restatement:] a public statement on AI extinction risk [added by the neutral restatement:] that referenced nuclear war, and [added by the neutral restatement:] each has separately [added by the neutral restatement:] stated a personal probability [added by the neutral restatement:] estimate in the 10 to 25 percent range for existential risk [added by the neutral restatement:] from AI, [added by the neutral restatement:] while both [added by the neutral restatement:] continue to [added by the neutral restatement:] lead companies building [added by the neutral restatement:] frontier AI.

Red-tinted words in the claim drifted from the evidence. Green-tinted words are what a neutral restatement needs.

The trace / claim to source

Where it appeared
misattribution
▲ Exaggeration
A real finding gets inflated: stronger, bigger, faster, or more certain than the evidence supports.
⌿ Omitted qualifier
A load-bearing condition from the source quietly disappears from the claim.
❝ Quote manipulation
Words attributed to someone that they did not say, or said with different meaning.
↺ Date or context mismatch
Real material from one time or place presented as another.
Tertiary sourceindividual commentary
Medium commentary by Kevin O'Shaughnessy, Nov 2025, quoting the Döpfner/Altman exchange in which Döpfner recalls Altman citing roughly 2 percent and Altman declines to treat it as a literal number
Tertiary sourceaggregator, authority unknown
calcuja.com P(doom) Survey 2026, characterizing the range as running "from Sam Altman's 'low but non-zero' to Dario Amodei's 10–25% range"
Tertiary sourceindividual commentary, included as an example of the same conflation the claim makes
James Pethokoukis newsletter, March 2025, asserting Altman, Amodei and Musk have all "admitted rather high estimates of p(doom)... about 10-25%"
Secondary sourcenamed publisher of the source interview
The Logan Bartlett Show newsletter recap of its own Dario Amodei interview, Oct 2023: "10-25% chance that disaster could occur"
Secondary sourcenamed-outlet journalism (event host)
Axios, Sept 17, 2025, reporting Amodei at the Axios AI+ DC Summit
Secondary sourcetabloid outlet, syndicated by Yahoo and AOL
TMZ, July 30, 2026, "OpenAI's Sam Altman Won't Put a Number on A.I. Wiping Out Humanity," with video of the exchange
Secondary sourcenamed-outlet journalism
TIME, April 13, 2026, recounting that Altman, Hassabis and Amodei signed the May 2023 statement, and separately noting Musk's 20 percent figure
Secondary sourcenamed commentator
Noah Smith newsletter, March 2026, stating Altman once told Döpfner he put extinction risk near 2 percent and later amended it to "big enough to take seriously"
Secondary sourcecommunity archive
LessWrong linkpost of Altman's 2015 blog post, "probably the greatest threat to the continued existence of humanity"
Primary sourceofficial body (statement operator)
CAIS "AI Extinction Statement Press Release," May 30, 2023, naming "CEOs of top AI labs: Sam Altman, Demis Hassabis, and Dario Amodei" as notable signatories
Primary sourceofficial body
CAIS signatory listing on safe.ai showing "Sam Altman, CEO, OpenAI" and "Dario Amodei, CEO, Anthropic"
Primary sourceAxel Springer produced interview
MD MEETS episode, Mathias Döpfner with Sam Altman, published Oct 2025, on Altman's 2015 blog post and superintelligence risk
● Primary source found
What is true
  • Both Sam Altman and Dario Amodei signed the CAIS Statement on AI Risk on May 30, 2023, confirmed by CAIS's own press release and signatory listing.
  • That statement does place extinction risk from AI in the same sentence as nuclear war and pandemics, as societal-scale risks warranting global priority.
  • Dario Amodei has repeatedly and on the record given a 10 to 25 percent figure, first on The Logan Bartlett Show in 2023 and again at 25 percent at the Axios summit in September 2025.
  • Both men continue to lead companies building frontier AI systems. Amodei has publicly defended continuing to build despite his own stated number.
  • Altman has publicly acknowledged existential risk from AI in strong terms, including in his 2015 blog post and in signing the 2023 statement.
What is misleading
  • Misattribution: the post says "Ten to twenty five percent by their own admission," directly after naming Altman and Amodei, which credits one man's number to two. The 10-25 percent range is Amodei's alone. No source found shows Altman ever stating it. The gap matters because the rhetorical force of the caption depends on both lab CEOs conceding the same high figure, which is the single most quotable element of the post and the one that is not supported.
  • Exaggeration: the claim states the two men "have separately given probability estimates of 10-25%." The evidence shows one man gave that range, and the other has been recorded at roughly 2 percent second-hand and has explicitly declined to state any number or range as recently as July 2026. Presenting one estimate as two converging estimates inflates the apparent consensus by an order of magnitude on Altman's side.
  • Omitted qualifier: Amodei's original phrasing was about something going "quite catastrophically wrong on the scale of human civilization," and his 2025 restatement was about things going "really, really badly." The claim converts both into "existential risk," a narrower and more dramatic category. Civilizational catastrophe and human extinction are not the same proposition, and the substitution is not neutral.
  • Quote manipulation: the CAIS statement says mitigating extinction risk should be a global priority "alongside other societal-scale risks such as pandemics and nuclear war." The post renders this as putting extinction risk "on par with nuclear war," dropping pandemics and converting a statement about policy priority into a statement about equal risk magnitude. The signatories did not assert that AI risk equals nuclear risk in size.
  • Date context mismatch: "by their own admission," in present tense, does not reflect Altman's most recent recorded position. In July 2026 he characterized the odds as "very low" and refused to give a figure, which is the opposite of admitting to 10-25 percent.
What is uncertain
  • I could not retrieve a transcript of the MD MEETS interview between Döpfner and Altman. The 2 percent figure reaches me only through commentary quoting that exchange, and even in that account the number is Döpfner's recollection of a private prior conversation rather than a figure Altman volunteered. I record it as reported, not established.
  • I could not view the TMZ video of the July 2026 Capitol Hill exchange. Three outlets carry the same text, but they are one reporting chain, since Yahoo and AOL are syndicating TMZ.
  • Proving that Altman has never uttered a 10-25 percent figure in any of hundreds of interviews is not possible. What I can say is that five distinct search strategies surfaced no such instance, and that the aggregators and journalists who catalogue these estimates place him well below Amodei.
  • The caption's motive attribution, that they kept building "because stopping hands it to a rival," was not traced to a specific quotation from either man in this investigation. Both have made race-dynamic arguments, but the phrasing is the poster's, not a sourced quote.
  • Whether the 62-video compilation the post is promoting represents these positions accurately was not investigated. Only the caption's factual claim was.
Evidence summary

On the signature. CAIS's own press release for the May 30, 2023 Statement on AI Risk names Sam Altman and Dario Amodei among its notable signatories, and both still appear on the CAIS signatory listing. The statement's full text is one sentence: mitigating the risk of extinction from AI should be a global priority "alongside other societal-scale risks such as pandemics and nuclear war." This half of the claim is directly confirmed by the primary source. On the 10-25 percent figure. The range traces to Dario Amodei alone. On The Logan Bartlett Show in 2023 he put the chance of something going "quite catastrophically wrong on the scale of human civilization" at somewhere between 10 and 25 percent. He restated the top of that range at the Axios AI+ DC Summit on Sept 17, 2025, saying there is a 25 percent chance things go "really, really badly," and reporting on a 2026 documentary describes him holding the same 10 to 25 percent range under questioning. On Sam Altman. Across multiple search strategies I found no instance of Altman giving a 10-25 percent estimate, or any range, for existential risk. The strongest numeric figure attributed to him is roughly 2 percent, and even that is second-hand: in the Döpfner interview, Döpfner recalls Altman citing about 2 percent in a prior conversation, and Altman responds by declining to treat it as a literal number while saying it is not zero and is big enough to take seriously. On July 30, 2026, asked point blank on Capitol Hill what percentage chance there is that AI wipes out humanity, Altman said the odds are "very low" and declined to give either a number or a range. An independent aggregator summarizing 2026 estimates characterizes his position as "low but non-zero" against Amodei's 10-25 percent. On "kept building." Both men lead frontier labs and continue to ship. Amodei has publicly defended building anyway, using an airline-safety analogy in a 2026 documentary interview. Altman's 2015 blog post called superhuman machine intelligence "probably the greatest threat to the continued existence of humanity," and he has since led OpenAI toward superintelligence explicitly. The "because stopping hands it to a rival" motive is a common paraphrase of both men's stated race-dynamic reasoning, but the post presents it as their own words without attribution to a specific statement.

Complete reasoning
As of 2026-09-06, the first half of the claim is confirmed by the primary source: CAIS's own records show both men signed the May 2023 statement, which does name nuclear war. The second half fails on its operative proposition. The 10-25 percent range belongs to Amodei alone, and Altman's recorded position is materially different: roughly 2 percent attributed to him second-hand, which he declined to treat as literal, and an explicit refusal in July 2026 to give any number or range. Because a source I cite bounds the claim away from Altman rather than merely simplifying it, the accurate family is closed, and "Mostly accurate" is unavailable. I rejected "False," since the signature element is fully verified and Amodei's figure is exactly as stated, so the evidence does not contradict the claim as a whole. I rejected "Source exists but framing is misleading," because the defect is not only framing: a specific numeric attribution to a specific named person is unsupported. Confidence is Medium rather than High because the disconfirming evidence on Altman rests partly on a single reporting chain and on an absence-of-evidence finding across five search strategies, and because I could not retrieve the Döpfner interview transcript directly.
Use this case

The reply is formatted for pasting into the thread where the claim is circulating.

Compact share page: ai.trueseeker.com/s/cf4e2b7e39d3/R4DFMQmOXdLGNQRUhjAQpgQboTj

Ask this case

Answers come only from the case file above; nothing is added.

Did Sam Altman and Dario Amodei both sign a statement comparing AI risk to nuclear war?

They both signed the CAIS Statement on AI Risk in May 2023, confirmed by CAIS's own press release and signatory list. The statement says extinction risk from AI should be a global priority alongside pandemics and nuclear war, but it does not claim the risks are equal in size.

Did both men really give a 10 to 25 percent chance of AI causing human extinction?

No. That range belongs to Dario Amodei alone, stated in a 2023 interview and repeated at 25 percent in September 2025. No source found shows Sam Altman ever giving that range or any similar figure.

What has Sam Altman actually said about the odds of AI causing extinction?

The only number tied to him is about 2 percent, and even that came second-hand from an interviewer recalling a prior conversation, which Altman declined to treat as literal. In July 2026 he was asked directly for a percentage and refused, calling the odds very low.

Is it accurate to say Amodei estimated existential risk specifically?

Not exactly. Amodei's own wording described something going quite catastrophically wrong on the scale of human civilization, or things going really, really badly. The post converts this into existential risk, which is a narrower and more dramatic claim than what he said.

Why does the fact check call this claim misleading rather than false?

The signing and the continued building are both confirmed true, but the post wrongly credits Altman with Amodei's 10 to 25 percent figure and reshapes the CAIS statement's wording, so the overall impression overstates the agreement between the two men.

Similar cases on record