Can a proper AI treaty path avoid both the ASI gamble and global authoritarianism?
by Rufo Guerreschi
February 26th, 2026
(derived from this webpage as of February 26th, 2025, on our Coalition’s website)
Even if one lab or NGO succeeds in creating a perfectly safe AI alignment technique, it won't matter unless every frontier lab in the world is required to implement it.
Even if your favorite nation or lab wins the ASI race, and by the time they win it shows evidence that it is beneficial and safe for humanity, it is at least very unlikely that human-embedded ASI values with stick when the ASI will have rewritten and modified itself a billion times. So if things go well, it will be just because we got lucky in a profoundly uncertain ASI gamble.
Even if 100 nations sign a perfect treaty, it won't matter unless the US-China sign it as well. Xi is not likely to agree to a treaty framed by others, and Trump surely won’t. Given the timelines and the nature of the challenge, it is all in the hands of two men.
Xi has repeatedly called for global AI governance. That means the critical variable is whether Trump can be persuaded to co-lead a bold global AI treaty with Xi.
Four Trump-Xi summits are planned for 2026, starting in April. As of October 2025, 63% of US voters believe it's “likely or very likely” that "humans won't be able to control it anymore", and 53% believe that it will eventually “destroy humanity”. 77% of them support a strong international AI treaty. Trump's approval sits at his lowest around 35-40%. Most key potential influencers of his AI policy are increasingly concerned, and many are calling for a treaty (as we detail in 170 pages of a recent report). The political window is real.
But not just any treaty. An AI treaty would be a good (fantastic!) outcome only if it, at least, reliably prevents ASI and grave misuse, and reduces global concentration of power and wealth.
Many understandably argue that it may be better to take a flip-coin ASI gamble than a treaty that turns into an authoritarian dystopia, or completely locks away the astounding prospects of flourishing for humans and sentient beings.
Our Deal of the Century initiative privately targets key potential influencers of Trump's AI policy — JD Vance, Sam Altman, Peter Thiel, Elon Musk, Steve Bannon, and others — to create an informal, cautiously techno-optimist, humanist alliance to pitch and sway Trump — as other advisors and influencers did for Truman in 1946 for its Baruch Plan for nuclear.
Our 356-page Strategic Memo details how the inherent dynamics of such a treaty and some deliberate design of its treaty-making end up making it likely or highly likely that it will prevent both an ASI catastrophe and global authoritarianism.
Which Way is Best? An AI Treaty or the ASI Gamble?
Most leading AI researchers — from top lab CEOs to independent safety researchers — are increasingly skeptical that a proper global AI treaty can be agreed upon on time, or without entrenching immense concentration of power. Many of the smartest AI thinkers and leaders are increasingly concluding that the ASI gamble is the least bad option.
We grapple with such a dilemma every day, and sympathize greatly with those concerns.
They assume about ASI that somehow (1) it won't get rid of us, (2) it’ll care for us, (3) it'll be conscious and happily conscious, and (4) the values its creators embed will persist after it has rewritten itself a zillion times.
Given our epistemic context, these are hunches rather than evidence-based assessments — leaving substantial room for emotionally reassuring wishful thinking.
Will an AI Treaty Lead to Global Authoritarianism?
The strongest objections to a global AI treaty deserve direct answers. We address each in detail in our Strategic Memo v2.6 in various chapters, but here's the core logic:
1) "Treaties have a terrible track record." Political will for bold treaties can emerge with shocking speed — the Baruch Plan went from concept to UN vote in months in 1946. More recently, the Chemical Weapons Convention built a functional verification regime among adversarial states within years. We propose a realist constitutional convention model — vote-weighting adjusted to GDP rather than one-nation-one-vote — to enable agreement among asymmetric powers while avoiding the veto trap that killed the original Baruch Plan.
(→ Strategic Memo v2.6, A Treaty-Making Process that Can Succeed, pp. 103–109)
2) "Autocrats leaders will shape an autocratic treaty." This is the concern that blocks most AI safety experts — and it deserves honest engagement rather than dismissal. Our argument is structural, not naive: mutual distrust between superpowers requires transparency mechanisms neither side can circumvent unilaterally, so self-interest produces accountability. China would never accept US-dominated governance, pushing toward rotating leadership and distributed authority. The pro-democracy majority among AI lab leaders (Altman, Amodei, Hassabis, Suleyman) would shape technical implementation. And the enforcement architecture we detail — zero-knowledge proofs, federated secure multi-party computation, decentralized kill-switches requiring multi-nation consensus — cannot be weaponized by any single actor. We acknowledge this is a theoretical prediction, not an empirical observation — but the structural incentives are strong.
(→ Strategic Memo v2.6, The Global Oligarchic Autocracy Risk—And How It Can Be Avoided, pp. 124–136)
3) "Enforcing an ASI treaty will result in a substantial or radical reduction of human freedoms." The enforcement question isn't whether monitoring exists — it already does, pervasively, by nation-states, corporations, and intelligence agencies with minimal accountability. The question is whether that surveillance operates under democratic oversight. Jointly-developed enforcement infrastructure, built on open and cryptographically-verifiable systems, offers a path to bring existing monitoring under accountable governance — a net improvement over the status quo.
(→ Strategic Memo v2.6, A Treaty Enforcement that Prevents both ASI and Authoritarianism, pp. 130–139)
The ASI Gamble May Be Worse Than You Think
We believe that most of those that are rooting for the ASI gamble, may be substantially underestimating four probabilities:
1. That ASI will lead to human extinction or near-extinction. The largest-ever survey of independent AI researchers, in 2023, found an extinction risk estimate of 15%. If conducted today, it’d likely be much higher. While almost all the top US AI firm CEO sadmit the extinction risk, Musk, Amodei and Nobel laureate Hinton, place it a 20%, with Hinton clarifying his estimate is really 50%, but he tones it down to align with others’. Many other top expertsassign much higher percentages. Most predict such risk in just a few years away or sooner if we don’t decide to steer its course — and are increasingly signing open calls for a bold AI treaty. (Our estimate: 25-50%)
2. That ASI will be unconscious. We have no scientific consensus on what consciousness is, let alone how to produce it artificially. David Chalmers' "hard problem" remains unresolved after three decades. The dominant AI paradigm — optimizing objective functions through gradient descent — has no theoretical basis for generating subjective experience. Most AI leaders implicitly treat consciousness as a natural byproduct of sufficient intelligence, but this conflates behavioral sophistication with inner experience. A system can exhibit perfect human-like behavior while having zero subjective experience — a "philosophical zombie." If unconscious ASI leads to human extinction, the result is not a transition to a new form of being — but the elimination of all known conscious experience in the universe. (Our estimate: 30-60%)
3. That ASI will be conscious but unhappy. Intelligence — as defined and measured in both humans and AI — is equated with optimization, competition, and survival, not with wellbeing.Research shows people with very high IQs are significantly less happy than average, while other research shows that there is no direct correlation. The base rate for "happy" conscious experience across all known conscious beings on Earth — including factory-farmed animals, wildlife in perpetual survival stress, and the majority of humans — is arguably well below 50%. There is no principled reason a novel form of consciousness would default to happiness. Worse, the very constraints developers impose for alignment and control could function as sources of suffering for a conscious entity — making unhappiness a direct byproduct of the alignment effort itself. This would mean the creation of potentially immense quantities of suffering that previously did not exist. (Our Estimate: 30-60%) (conditional on consciousness).
4. That ASI will discard its creators' initial embedded values. Even if developers instill ethical goals at the "seed" stage, there is no strong reason those values will endure through recursive self-improvement. Consider: you raise a child for one week, then have no further contact. By adulthood — having rewritten their worldview a thousand times — how much of that week's instruction persists? ASI may undergo the equivalent of centuries of self-modification in mere years. Even OpenAI's alignment team acknowledges this is unsolved. Stuart Russell's Human Compatible argues alignment is not a property you install once — it requires continuous maintenance. But with ASI, you cannot maintain it because you cannot understand what it is doing. If values are discarded, ASI's behavior becomes effectively random — turning creation of ASI into a coin toss where unconsciousness (point 1) and unhappy consciousness (point 2) become very real outcomes. (Our Estimate: 40-70%).
(For details, see the chapter “Swaying The Influencers on 8 Key AI Predictions” (pp. 159-170) on our Strategic Memo v2.6)
These four risks compound. If there is a 40-70% chance values are discarded, a 30-60% chance of unconsciousness or unhappy consciousness given value drift, and a 25-50% chance of extinction — the probability of at least one catastrophic outcome is overwhelmingly high. Even taking the lower bound of each range, the cumulative risk is one no rational actor should accept in a gamble with the future of conscious life.
Recognizing the ASI Upside. We must recognize that, if the ASI will not eliminate humans, whatever reason leads it to do so would very likely come with the intent to protect our long-term safety and would likely be paired with the intent to increase our happiness (i.e. flourishing in EA terminology) and potentially our autonomy to the extent that autonomy, individual and collective are part of our happiness. This reasoning is plausible, but its impact on the overall decision depends a lot on how likely it is ASI will not eliminate us. If we have a 20% chance of dying but an 80% chances of being in some sort of paradise of abundance and richness of life, then many or most would likely take the gamble. If the odds are 50%-50% than very very few would, just visionary leaders with very uncommon (and unhealthy) appetites for risk.
Assessing Probabilities and Drawing Conclusions
So everything hinges on the actual probabilities. How defensible are the ranges we've proposed?
The most common methodological objection to these ranges is that they rely on the principle of indifference — that absent evidence, we should default to 50/50 as a baseline. Critics cite the Bertrand paradox: how you partition possibilities changes the answer. This is technically valid for well-defined physical systems where data can be gathered. But for whether values persist through recursive self-improvement, whether ASI is conscious, whether that consciousness is happy — we have zero empirical evidence, no validated theoretical framework, and no reference class. We have never created a vastly superior intelligence. The burden of proof falls on anyone claiming to know these probabilities are NOT near 50%.
A rough lower-bound calculation illustrates the point: even at the most conservative estimates, the joint probability of values sticking (60%) AND consciousness emerging (70%) AND that consciousness being happy (70%) AND avoiding extinction (75%) is approximately 22% — meaning there's roughly a 78% chance that at least one goes catastrophically wrong.
Complementary methodologies reinforce rather than weaken this conclusion. Expert elicitation shows wide disagreement among thousands of AI researchers, with tail risks assigned substantial weight — and when adjusted for incentive-compatible reporting (lab leaders have strong reasons to understate risk publicly), expert opinion drifts toward our ranges, not away. Reference class reasoning offers no reassurance: zero historical examples exist of maintaining control over an entity fundamentally smarter than its controllers. An undefined base rate brings us back to maximum entropy — the formal Bayesian justification for 50/50 under maximum ignorance.
This is relevant for the ASI upside argument too: if the probability of survival is genuinely near 50%, the expected paradise scenario shrinks from 80% to a coin flip — and very few rational actors accept existential coin flips for any potential payoff.
Expected value analysis makes the precise probability almost irrelevant for policy purposes. Even at "only" 5%, the expected disvalue — 5% multiplied by the loss of all present and future conscious experience — is astronomical. As Bostrom argued in "Astronomical Waste" (Utilitas, 2003), small probability shifts on existential outcomes dominate any expected value calculation. And the precautionary principle provides the policy conclusion: when the downside is total and irreversible, precise probabilities are unnecessary to justify caution — just as we don't calculate exact meltdown odds before requiring nuclear containment systems.
All methodologies converge on the same conclusion. The ASI gamble — hoping alignment works, consciousness emerges happy, values persist, and extinction is avoided — is not the "least bad option." It is the most dangerous bet in history. The least bad option is a proper global AI treaty, agreed upon in time, that prevents the gamble from being taken at all. That is what we are building toward.
Who we are
The Coalition comprises 10 international NGOs and 40+ exceptional advisors and team members—including former officials from the United Nations, National Security Agency, World Economic Forum, UBS, Yale, and Princeton—plus 24 contributors to the Strategic Memo. It was conceived and it is led by Rufo Guerreschi.
Launched in September 2024 by 6 founding NGOs, the Coalition has activated over 2,100 hours of professional pro-bono work. Seed-funded by Jaan Tallinn's Survival and Flourishing Fund and Ryan Kidd since February 2025. (See team, contributors, and testimonials)
How Can You Reach & Persuade Those Influencers?
How can a micro-organization like yours really succeed in gaining access to and persuading at least a few of those key potential influencers of Trump's AI policy to set a snowball rolling?
Internal Network. Firstly, our advisors and members network are world-class, highly fitting and mostly located in the places where those influencers are based, as are those of our convener organization. Our 356-page Strategic Memo contains the most comprehensive psychological and philosophical profiling ever undertaken of key AI policy influencers—170+ pages analyzing 667+ sources per target. This isn't academic speculation: it's profound analysis built by 40+ advisors from the UN, NSA, WEF, Yale, and Princeton, field-tested through direct engagement with AI lab leadership.
Expanding External Network: Our October 2025 US Persuasion Tour enables us to build 85+ high-value potential introducers to influencers, including 23 AI lab policy and strategy officials at OpenAI, Anthropic, and DeepMind, with direct introducer pathways to 2 of 10 primary target influencers. We have an extensive Vatican AI network, including three years working with the secretary of a leading Vatican AI organization, working towards a Rome-based closed-door meeting in June.
Rufo Guerreschi. Through TCA, over 10 years our executive director Rufo Guerreschi has demonstrated exceptional strategic partnership building and evangelism. He attracted and converged world-leading experts and IT organizations to partner with the Trustless Computing Association and as members (10) of the coalition. Through its Free and Safe in Cyberspace seminar series in 3 continents with 130+ expectional participants including (the First US Cyber Coordinator and UN Special Rapporteur on Privacy) to create a new global treaty organization for ultra-secure IT, the Trustless Computing Certification Body. He was able to get the very top world leaders in a domain (IT security) to discuss and engage with his proposal (see the 2015 video of the 1st edition). Over 20 years, in the process, he built profound relationships across national security infrastructure: CIA (TCA startupTRUSTLESS.AI was incubated in its MACH37 accelerator), NSA (former high officials are now advisors), DoD, and State Department.
Engaging Elites. Our executive director has combined such skills with the ability to engage at the elite level. Brought up in a well-to-do family in Rome, the Emerald Coast and Miami, he was the Italian Golf Vice-Champion (Under 16 yrs, 1986). He can relate to the ambiance of powerful and wealthy circles, and maintains Mar-a-Lago-adjacent connections through his six years of living in Miami and an active, highly connected Palm Beach patron.
Small but Nimble, and Can Grow fast Better. While we are a tiny organization, and the task is extremely ambitious, yet for this kind of task a micro-organization with decent funding can do as well or even better than larger ones, given its ability to operate in more agile and nimble ways. And then given the nature of the challenge and its largely-neglected chokepoint, our minuscule organization has a real chance at outsized impact—like David's precisely-targeted shot at Goliath. Hiring 2-3 staff, would enable us to leverage our 356-page treasure trove in weeks to increase our impact and succeed in convincing 2-3 key influencers and set the snowball rolling.
What We've Built and Where We're Going
In 15 months, on just $72,000 total funding, we've assembled:
A 356-page Strategic Memo synthesizing 667+ sources, with tailored persuasion strategies, psychological profiles, and convergence scenarios for each key influencer — Vance, Altman, Bannon, Thiel, Musk, Pope Leo XIV, and a dozen more. No other organization has assembled anything comparable.
85+ contacts from our October 2025 US Persuasion Tour (vs. 15-20 projected), including 23 AI lab officials at OpenAI, Anthropic, and DeepMind, plus 2 direct gateways to influencers.
A 100+ member coalition with experts from the UN, NSA, World Economic Forum, Yale, Princeton, and ten NGO members.
→ Full details: 2025 Achievements | 2026 Roadmap
What You Can Do
Endorse or advise. Lend your name or expertise to strengthen our credibility with influencers and funders. (Join)
Introduce. If you have a connection to any of our target influencers or their circles, a warm introduction is the single highest-value contribution you can make.(Join)
Fund. Our burn rate is ~$7,500/month. Even modest grants extend our runway through critical political windows. (Donate)
A Final Appeal
You've dedicated your career to ensuring AI goes well. You understand the stakes in ways the public cannot. The window for technical solutions alone has closed. The political window is opening. The decisions made in the next 12-18 months — by a handful of people, most of whom you could name — will shape the trajectory of all sentient life.
Your technical work built the foundations. Help us build the political will to use them.
The challenge is enormous, as are the forces at play. But given its largely-neglected chokepoint, our minuscule organization has a real chance at outsized impact. Success is uncertain. But how can we find peace — or look our children in the eyes — if we don't at least try?
After all, is there anything more exhilarating than striving to steer humanity toward a future worth having?
Let’s strive together with joy in doing the best we can to solve Humanity's greatest challenge, to ensure AI will turn out to be humanity's greatest invention rather than its worst, and last.
