An Alignment of the Vatican and Anthropic for a Positive AI Outcome?
On May 25th, Pope Leo XIV will present his long-awaited encyclical on AI, Magnifica Humanitas, in an in-person event in Rome. An AP wire headlined the event “Pope and Anthropic co-founder to launch pontiff's AI encyclical on May 25th,” a framing picked up by the Washington Post, PBS, and Bloomberg.
While the Vatican's own press release effectively names Olah as one of only five opening panelists, the headlines substantially overstated the implications of Olah's speaking role. Despite the diplomatic stakes, neither institution has taken any action to deflate the media compression, and their silence is a signal in and of itself.
Given current AI development timelines, the chances of staving off the immense safety and power-concentration risks of AI rest almost entirely on whether Trump will choose to co-lead a bold treaty with Xi.
Both institutions, among the best positioned to catalyze that choice, were temporarily sidelined by sympathizers and operatives of Peter Thiel's post-humanist AI agenda — which has so far dominated US AI policy with its hands-off stance to the immense gamble of the ongoing and completely ungoverned race to Superintelligence, or ASI.
Yet that agenda is ever less popular — among US voters, and inside and around the administration. On May 25, the two institutions will share a crucial platform. Can the Vatican and Anthropic play a decisive role in catalyzing a critical mass of key potential Trump AI policy influencers on a Deal of the Century for AI?
Anthropic, Global AI Governance, and the Vatican
In the AI race, Anthropic is head-to-head with Google and OpenAI and gaining ground. While being the fastest-rising US AI lab in terms of revenue (up 10x year-over-year), capabilities, and a $900 billion valuation (up from $160 billion six months ago), its leaders have been the loudest about the immediacy and scope of both the power concentration and extinction risks of AI.
For these reasons, Anthropic was singled out as a woke AI and as regulatory-capturing “doomers” by the (now former) AI Czar David Sacks and more recently attacked by the Secretary of War Hegseth for its requested terms of use. Meanwhile, its latest AI model preview, Mythos, shocked the world with its defensive and offensive cybersecurity capabilities.
As it became clear that Anthropic had become the preferred AI for the most advanced US military and intelligence operations, direct engagements at the highest level of the White House followed, which largely repaired the relationship. Even Trump put in a good word, and the ability of the US government to use Anthropic's AI was restored, even while the court case with the Pentagon continues.
For its Claude Constitution — the document that Anthropic hopes its AI model will learn to obey before it is too late — Christian and Catholic leaders were engaged, including Msgr. Paul Tighe, co-author of the Vatican's main AI doctrine document. Discussion included, among other things, whether Claude could be considered “a child of God” — an idea in line with Anthropic leadership, who mostly think it is probable or highly probable that current and future AI are conscious. That's likely how the atheist Christopher Olah became a key external AI reference for the Vatican.
The Vatican's position on AI Governance
The Vatican has been forcefully positioning itself on AI ethics and governance. Pope Leo XIV leads 1.4 billion Catholics in the world, including 20% of US citizens, and is looked up to by most US Christians, with the highest favorability of any leader among US citizens.
Pope Leo XIV has placed AI unequivocally at the center of his papacy, favoring regulation for AI ethics and preventing concentrations of power. His predecessor, Pope Francis, called for a binding treaty, and the current pope's most recognized Vatican AI expert, Fr. Paolo Benanti led in September a coalition of top scientists calling for a bold treaty to ban Superintelligence.
Notably, however, the Pope has so far stopped short of ANY explicit language on loss of control, human replacement, or extinction risks — and its key AI officials have mentioned it relatively very rarely in their documents and articles. This is a critical, vital, and existential gap that will hopefully be decisively closed via the May 25 encyclical.
The asymmetry is glaring: Anthropic's own leaders describe a shocking near-term existential risk. Given the solid scientific plausibility of very short ASI timelines, the Vatican's silence on loss of control risk would be a hugely consequential category error. If the Vatican took Anthropic's published estimates seriously, it would not be cultivating moral influence over AI training corpora — it would be running an emergency campaign for a US–China-led AI treaty. Yet, both speeches of May 25 and their follow-ups are opportunities to close such a gap.
The Dominant US AI Policy So Far
While the Pope is yet to shape definitively his view on AI via the encyclical, the Vatican's position on global governance of AI — albeit timidly expressed so far — is in direct contrast with the current, increasingly minoritarian US AI policydominated by Peter Thiel, and his uniquely powerful AI network inside the Trump administration. Such a policy, supported by very solid secular concerns and likely-heretical theological arguments, is centered on a completely hands-off approach to the race to Superintelligence and AI concentration of power.
That's why Thiel was not happy at all when, in May 2025, JD Vance stated that the US should defer to the Pope on AI ethics and safety. It was starkly criticized by Peter Thiel a few months later as something that could create a “Caesar-Papist fusion” on AI that could globally affirm a vision contrary to his. This is the sole reason Peter Thiel even came to Rome last March, a few steps from the Vatican, for 4 days of lectures. His stated aim was to argue that those that will lead a movement to create a global treaty to prevent the loss of control and extinction risks of AI will lead inevitably to a global authoritarian dictatorship and represent the Antichrist — something he has repeated ever more forcefully since last September — and pre-emptively persuade as many Vatican power brokers as possible of that.
Thiel’s feared risk of a “Caesar-Papist fusion” on AI was further deflected when Vance criticized the Pope on his theological competency on just war when Trump rebuked decisively on the Pope’s criticism of the US decision to wage war on Iran (except that Marco Rubio, a cradle Catholic increasingly considered more likely than Vance to succeed Trump as the 2028 presidential candidate, came to visit the Pope a few weeks ago …).
Anthropic's Positions on AI Governance
Anthropic’s vision for global governance of AI is mostly shaped by Dario Amodei, followed by his sister, co-founder and Anthropic President Daniela, and her husband, Holden Karnofsky, then co-founder Jack Clark, and then other top officials and board members.
Dario Amodei stated in 2023 there is a 10-25% risk that AI could destroy humanity (i.e., extinction) and that we are months or maybe years away from AI capable of being “a million geniuses in a data center” (i.e., AI that can self-improve at an ever-accelerating rate, with extreme risk of loss of control). Last September, he stated, “I think there's a 25% chance that things go really, really badly” — i.e., human extinction or equivalent.
While he is wholeheartedly in favor of global AI coordination to prevent loss of control and other misuse and safety risks, he has been equally concerned by the China AI authoritarian threat and largely skeptical of most of the proposals — as per a recent interview with the New York Times’s Ross Douthat. From minute 37.54, he said that if a US and China treaty to enforce safety constraints and slow down were really possible, he'd be “all for it,” but that they would be technically and politically possible when and if there will be “truly reliable verifications.” From minute 40.02, he said he was “extremely skeptical” of a US-China treaty to prevent firms and states from pursuing ever more powerful AI because of the current scale of the economics behind the race and the benefits of AI. Further along, from minute 48.35, he stated that a treaty limited to banning some architectures or techniques, including AIs that continually learn and implicitly recursive self-improvement, would still see him “very skeptical, but at least it is not dead on arrival.”
Jack Clark, co-founder and global head of policy until a few weeks ago, has been the strongest voice among US AI leaders (after perhaps Amodei and Musk) in warning of enormous and imminent risks of full recursive self-improvement (i.e. loss of human control over AI), even producing in May 2026 a scientifically detailed case that its odds are 30% by 2027 and 60% by 2028. He is a longtimesupporter of strong global AI safety governance, even supporting literally the Baruch Plan as a model for AI governance, and now leads the new Anthropic Institute. Unsurprisingly, he was singled out for accusations of being a “woke doomer” by the (now former) White House AI Czar Sacks — who is deeply aligned with Thiel (while paradoxically stating once he “thinks all the time” about loss of control).
Anthropic Strategic Hesitation between Two Immense AI Risks
With predictions so stark by Amodei and Clark about the risks of loss of control and extinction prediction, you would expect they'd be extremely loudly and unequivocally calling for a bold and stringent AI treaty. But, like Thiel and many other AI lab leaders, they are also greatly concerned about (1) the actual ability of a global AI treaty to reliably and durably ban ASI and (2) inadvertently ushering in durable forms of global authoritarianism — and more so is Holden Karnofsky.
As husband of co-founder and President Daniela Amodei, former long-time director of the massively influential Open Philanthropy (now Coefficient Giving), and a de facto strategic co-leader, Karnofskyemphasizes that Anthropic should prioritize the risk of a human “power grab” over extinction risk. In a recent interview on 80,000 Hours, he acknowledges the risk of AI misalignment (i.e., loss of control or extinction) and (human) “power grabs” are “maybe in the ballpark of each other in terms of how important they are.” However, he expressed concern that the current community is “a little overfocused on the misalignment risk,” concluding, “Probably misalignment risk is a bigger deal in my opinion, but they’re close, so I think we could use more attention on the power grabs.”
Interestingly, Karnofsky's stance aligns directionally with the core stated position of Peter Thiel, though Thiel's focus has been primarily on warning that leaders of movements for a global AI safety treaty are willingly or inadvertently abetting a global authoritarian regime.
In the same interview, Karnofsky states he believes most lab leaders are already committed to pursuing ASI and are not seriously seeking to halt or slow development. And that is arguably, at least in part, for the same reasons he does. At minute 3.32.20 he states that, once AI R&D becomes completely automated,he and a few of Anthropic AI leaders will command an army of “AI researchers” to “move into the open-ended” area (i.e., recursive self-improvement and intelligence explosion).
Unlike Amodei and Clark, he seems at peace with such a scenario, in the scheme of his assessment of competing risks. Given its effective altruist frame of mind, such comfort implies he believes with a higher probability than evidence warrants that the resulting ASI will:
(1) retain any of the values instilled by its designers after it has rewritten itself a million times in weeks or months;
(2) not result in human extinction or human dystopia;
(3) have some form of consciousness rather than being a mindless, ever-expanding digital cancer; and
(4) not result in a suffering or miserable conscious entity.
In addition to such (arguably) largely unwarranted probability assessment, Anthropic's stance is also connected to the idea that ASI may be a worthy successor, if humanity does not make it and, more generally, a species-agnostic ethics aimed at the well-being of future conscious beings, a philosophical stance increasingly influential at Anthropic, and most other AI lab leaders, but one Anthropic cannot publicly defend in a Vatican setting. No Christian tradition holds that an artificial mind, however well-tuned, will be conscious or inherit the imago Dei.
As odd as it may sound, an intense and explicit dialogue will be needed to arrive at shared, probabilistic and explicit determinations about such deeply-uncertain ASI predictions — not only between Anthropic and the Vatican but also among other key actors. Still, as hugely important as it is to jointly assess the likelihood that AIs will or should be moral patients, we are faced with more important and shockingly consequential AI challenges.
The Prospects of Convergence
The fact that a global AI treaty could turn into a global tyranny — Thiel's and Karnofsky's key concerns — is a real and under-appreciated risk, one deeply shared by top AI CEOs. Yet, they overlook that NOT having a treaty will default to such global tyranny — if humanity remains in control and survives.
From that it derives that preventing an immense and durable concentration of power — by individuals who might use AI to radically extend their lifespans, perhaps indefinitely in digital or biological form — requires exactly the same enforceable global treaty needed to prevent loss of control.
Both require a proper, timely, and bold treaty. There is no version of “stop the power grab” that does not also stop the runaway race, and no version of “stop the runaway race” that does not also stop the power grab.
If ASI will lose all its embedded values as it rewrites at an ever-accelerating rate — as it is very likely — none of the Catholic ethics that the Vatican may hope to embed in Claude's Constitution, and those of other leading AIs, will survive beyond an ASI's first few weeks or months.
A structural alignment of Anthropic and the Vatican towards an actionable path that prevents both risks is likely, and more so if (1) Anthropic helps the Vatican to better understand the loss of control risks and (2) Anthropic becomes more optimistic about the prospect of a proper AI treaty, as it sees the emergence of a Vatican-backed vision for a treaty-making process shared with a critical mass of other key potential influencers of Trump's AI policy.
Among those, top AI lab leaders (like Hassabis, Altman, Musk, and Suleyman) and key political and media leaders like Bannon (already strongly in favor of banning ASI and an “immediate treaty”), Glenn Beck, Joe Rogan, Tucker Carlson, and Ron DeSantis — are increasingly and loudly calling out AI's safety and power-concentration risks.
By now, 63% of US voters believe it's likely that “humans won't be able to control it anymore,” 53% believe “AI will destroy humanity” is somewhat or very likely, and 77% of them support a strong international AI treaty.
The White House's rift with Anthropic is mostly amended, and the AI Czar that attacked it is no longer in place. While Trump's clash with the Pope on the Iran War was harsh, it was not about AI — and Rubio, Vance and DeSantis are all more or less devout Catholics. Furthermore, scientists, citizen surveys, and MAGA leaders are all shifting towards heightened concerns about AI, a treaty with China, and some kind of regulation within the US.
A Catalyzing Role for the Vatican and Anthropic?
Given all of the above, it is increasingly plausible that Anthropic and the Vatican could be part of a broader informal alliance or coordination of key potential influencers of Trump's AI policy that will aim and succeed in moving Trump from no AI regulation to a proper global and domestic framework — and then support him in advancing such a new policy.
Facilitating exactly this kind of convergence — for a treaty that will resolve Amodei's concern for 'truly reliable verifications' and mitigate sufficiently the risk of power concentration highlighted by Karnofsky and Thiel, among other risks — is the working hypothesis of the Deal of the Century, led by the author.
Getting Secretary of State Marco Rubio, the son of Cuban immigrants, to co-lead could become key to swaying Trump. For these reasons, it's worth noting that Cardinal Czerny, newly appointed to coordinate the Vatican's new AI commission, has also publicly and candidly assessed the morality of Trump's Cuba policies in recent weeks — a portfolio overlap the Vatican may want to disentangle if it wants the AI conversation to stay constructive with the current US administration.
Conclusions
Olah's speech on May 25th could be one of the most consequential moments of 2026 for AI governance, even if the wire services have not yet noticed.
If Anthropic leadership, through Olah's speech, decides to use the platform to put on the record the shocking loss-of-control estimates from Anthropic's own leaders and frame these as the moral context for the encyclical's reception, it will have implicitly closed the Pope's vocabulary gap on loss of control without forcing the Pope to do so himself. That alone would reframe the encyclical from a document about dignity and labor into the founding text of a forceful and decisive Catholic case for a proper and urgent AI treaty.
If he does not, the encyclical will be remembered as the Vatican's strongest ethical statement on AI, as well as a missed historic opportunity, unseen since 1946, to assist humanity in developing the global institutions required to address inherently global challenges.
Pope Leo and Olah can point to the single solution that defuses both risks at once: a bold, timely, and proper US–China-led treaty-making process with the Vatican as moral guarantor and Anthropic as technical and economic validator — for Trump and a critical mass of converging key potential influencers of his AI policy.
This is the convergence of the Deal of the Century initiative — led by a Rome- and DC-based coalition of 10 small international NGOs and directed by the author — was designed to facilitate, and which its first Open Letter to President Trump, routed to the White House on May 12, set in motion.
As the leaders of all 4 front-running AI labs — Musk, Amodei, Hassabis and Altman — have all more or less explicitly stated that we are on the edge of, or even beyond, a point of no return in the race to Superintelligence, it is impossible to overstate the urgency with which institutions like the Vatican should act — while there is still agency.
