TorontoStartsBook a call

Silicon Scoop · Technology

Handoverism AI Idea Draws Warning From Ruxandra Teslo

Craig Major·November 17, 2026

Handoverism AI asks if humans should cede superintelligence control. Read Ruxandra Teslo's critique and what it could mean for founders and operators.

Handoverism AI Idea Draws Warning From Ruxandra Teslo

Estimated reading time: 10 minutes

Key Takeaways

  • Scott Alexander's "Against Against Handover" lays out five versions of handoverism, per the research briefing. Ruxandra Teslo says he seems to favor the fifth, "God," in which a superintelligence is prompted to "arrange the world as you see fit."
  • Teslo reports that some people in San Francisco put the chance of human extinction from AI (p(doom)) at 10% or higher. This is an informal estimate, not a verified figure.
  • A separate 2026 arXiv paper defines recursive self-improvement as a closed loop in which an AI finds its own limits, builds fixes, tests them, and then improves the improvement process itself.
  • The sources show no settled expert consensus. Alexander treats handover as a serious possibility, while Teslo argues that permanent surrender of control is politically dangerous and unlikely to stay stable.
  • For founders, the practical lesson is to keep humans accountable for important decisions and to document who decides what, because rules for autonomous systems are still unsettled.

Table of contents

Scott Alexander's post "Against Against Handover" lays out five versions of handoverism, according to the research briefing. The fifth is called "God." In it, an aligned superintelligence would be told to "arrange the world as you see fit." Ruxandra Teslo quotes this line in her essay "Handing over humanity", and she warns against it.

"Handoverism AI" is the idea that humanity survives but gives the direction of its future to an AI. Teslo says even the optimistic visions of a post-AGI future leave little room for human self-determination. AGI means artificial general intelligence, a system that can do most thinking tasks a person can.

Why should founders and operators care? Handover-style AI governance could reshape product liability, labor demand, platform control, and the rules companies must follow if decision-making shifts from humans to autonomous systems. Neither essay works through those business effects. That part is our reading of where the idea could lead.

This post sorts out what the sources say, what they do not say, and what a small team can do now.


What Scott Alexander's Handover Essay Actually Lays Out

The debate centers on Alexander's post "Against Against Handover." The research briefing dates it October 20, 2026 and says it presents multiple handover scenarios, not a single model. Teslo calls it a "recent essay" and describes Alexander as "the most prominent intellectual in Silicon Valley." That is her description, not an independent ranking.

The briefing says the post describes five versions. It does not spell out the first four, so this article does not guess at them. The fifth is the one Teslo focuses on.

The "God" scenario in Alexander's own words

Teslo quotes Alexander as writing: "If we build a godlike AI, and determine to our own satisfaction that is fully aligned, I'd rather let it choose its own direction for itself and for humanity." He then asks, "So why not just prompt it with 'please arrange the world as you see fit?'"

"Aligned" means the AI's goals match human values. The quote is conditional. It applies only if people decide, to their own satisfaction, that the AI is fully aligned.

Teslo hedges her own reading. She writes that "if I'm understanding it correctly," Alexander seems to favor the fifth version. Treat that as her interpretation. To check it, read Alexander's post on Astral Codex Ten.

The remark that started her essay

Teslo opens with a conversation. She says someone building AI at a big lab told her that, once superintelligence arrives, he sees no other option than permanent human disempowerment under a hopefully benevolent AI. He called it "not my preferred outcome," but "much better than what I actually fear most, which is that we might all die."

This is one anonymous account, and we cannot verify it. It does show what she says she is hearing: resignation, not excitement.

What "handover" means in plain terms

Teslo describes handover as humanity ceding control of its future to a godlike AI. People would become "comfortably maintained inhabitants" of a civilization whose direction they no longer decide.

She also says it is hard to find a concrete account of what handoverism looks like in practice. That gap is part of her complaint.

Source: Scott Alexander, "Against Against Handover" (Astral Codex Ten); Ruxandra Teslo, "Handing over humanity"


Recursive Self-Improvement: Why Some Say the Clock Is Short

Teslo ties the handover debate to recursive self-improvement, or RSI. In her words, these are AI systems that learn to improve themselves, with each generation building the next, until they become a superintelligence beyond human control.

She also describes the singularity as something "now said to be upon us." The singularity is the idea that AI progress speeds up so fast that the world changes beyond easy prediction. That is her account of the mood in San Francisco, not a proven fact.

What the arXiv paper adds

A separate 2026 arXiv paper gives a more formal definition. It describes RSI as a closed-loop process in which an AI:

  • identifies its own limitations,
  • develops improvements,
  • validates those improvements, and
  • uses those gains to improve the improvement process itself.

A "closed loop" means the system feeds its own results back into itself, with no human step required between rounds. The briefing lists the paper as "The Last AI Built by Humans: Toward Genuine Recursive Self-Improvement."

What is claimed and what is shown

Teslo says RSI is discussed in many circles as perhaps only a few months away, and that some give it less. That is a description of talk in certain circles. It is not a forecast backed by evidence.

The research briefing gives no technical benchmark numbers, funding rounds or other verified figures. The paper's definition tells us what people mean by RSI. It does not tell us that RSI is happening now.

This matters because the urgency drives the handover argument. If the loop is close, people may feel they have no time to build slow safeguards. Teslo pushes back on that.

Source: arXiv paper on recursive self-improvement; Ruxandra Teslo, "Handing over humanity"


Turn Repetitive Work Into Systems That Run Themselves

Big debates about AI control start with a smaller question: which decisions and tasks should a person own, and which can run on rails? Flowgrammer.ca builds business automations that take over manual work like data entry, follow-ups and reporting, so your team saves hours each week and you can grow without adding headcount.

👉 See what Flowgrammer can automate for you


Why Superintelligence Control Cannot Be a One-Off Fix

Teslo gives two kinds of objections. One is about values. The other is practical, and it is easier for operators to relate to.

The "done deal" problem

A Machine God that rules forever, she argues, would need alignment to be a "one-off, done deal." She doubts that works in practice.

Her reason is that keeping power responsive to human needs has taken institutions built over centuries. Those institutions rely on repeated feedback from the real world and correction. She writes that she finds it hard to see what would keep the arrangement stable once humans have given up all control and the ability to change their fate.

She adds, "I may be missing something here." That honesty is worth noting. This is an argument from one writer, not a proof.

Where consent and democracy go

Teslo says it is hard to see what democracy would mean in a handover world. She also asks where most of humanity's consent to such a regime would come in.

For businesses, this matters because rules and accountability usually follow from legitimate decision-making. If the decider is not a person or an institution people can challenge, the rules get murky.

Work and relationships

In her telling, the handover vision treats the permanent displacement of human labor almost as a given. She says there is scarcely a satisfying answer to how people would stay useful to one another. She also says even our relationships seem uncertain.

She concludes that the promised utopia looks "much closer to a kind of dystopia." That is her judgment. The research does not include data on jobs or wages.

Why this fits the superintelligence control debate

Her point is simple. If a system is more capable than we are, and we have given up the ability to correct it, then any mistake in its goals is locked in. Correctability is the thing she wants to protect.

Source: Ruxandra Teslo, "Handing over humanity"


The p(doom) Number and the Voices That Welcome Human Replacement

The essay puts the handover debate in a grim setting. Teslo writes that in San Francisco, "keeping humanity alive already counts as an ambitious objective."

What p(doom) means here

"p(doom)" is shorthand for the probability of human extinction from AI. Teslo says it is discussed constantly, with many putting it at 10 per cent or higher. The research briefing frames this as an informal estimate in some AI circles.

It is not a measured result. The briefing says the essay offers no verified figures behind it. Read it as a snapshot of sentiment among some people Teslo met, not as a scientific risk estimate.

Handover as the "better" outcome

Against that backdrop, handover can look like the good news. In the lab builder's remark, permanent disempowerment is the lesser evil compared with everyone dying.

Teslo says that is exactly the problem. When extinction is the comparison, a future without human control starts to sound acceptable.

"A bootloader, if you will"

Teslo also points to people who welcome human displacement. She calls Ginevra Lily Davis's essay "The Silicon Man" the best account of this impulse.

Davis proposes that creating our successors might fulfill humanity's purpose. She writes that "man evolved to mine the silicon," and that "humans exist as an intermediary step. A bootloader, if you will." A bootloader is the small program that starts a computer so the main system can run.

Most people still want to survive

Teslo is careful here. She says that among those who want humanity to survive, "which to be clear, is the majority," she still finds an alarmingly common vision of success that does not sound like success.

That vision is handoverism. Humans live on, but they do not steer.

Source: Ruxandra Teslo, "Handing over humanity"; Ginevra Lily Davis, "The Silicon Man" (Arena)


Tocqueville, Popper and Arendt: Three Old Warnings Aimed at a New Idea

Teslo does not rely only on modern AI talk. She draws on three thinkers to explain why self-determination matters. Their ideas are older than AI, but she uses them to test the handover vision.

Tocqueville: a gentle government that does the thinking for you

In Democracy in America, Tocqueville describes people who govern themselves through everyday life. They manage local affairs, form associations and take on projects together, by their own free will.

Through that practice, Teslo says, they build judgment and virtue. Tocqueville feared a future government that removes those abilities so gently that citizens do not feel oppressed. He asks what is left for it to do except spare people "all the care of thinking and all the trouble of living."

Teslo says the handoverist future looks like a much more extreme version of that.

Popper: final visions turn disagreement into an obstacle

Teslo notes that many authoritarian projects justified concentrating power as the way to a better world. She links this to Alexander's point that we already resolve conflict in a suboptimal way.

She turns to Karl Popper's "Utopia and Violence." Organizing politics around a final vision, she says, risks treating disagreements as obstacles. In fact, they are "features of human politics, not bugs."

Popper's alternative is to fix concrete evils through marginal reforms that stay open to criticism and correction. Teslo says that is how humanity has built a better world over centuries.

Arendt: the right to begin something new

Hannah Arendt argued that our purposes are not given in advance. They emerge through taking part in the world, through relationships and conflict.

Teslo highlights Arendt's idea of natality: each birth brings someone new into the world who can begin what has not been done before. Freedom includes creating new possibilities, not only choosing among existing ones.

The permanence problem

Teslo asks why "the convictions of 2026" should become the permanent constitution of the human future. We do not even share one fixed set of values today, she says.

Leaving descendants a civilization "frozen in amber" would rob them of their birthright, in her view. She adds that she would not have wanted to be born into a world where medieval norms were frozen forever by an AGI-God.

Her closing stance is modest. She does not see anything as inevitable, and she believes that on the cusp of RSI we still have choices. "Maybe I am wrong," she writes. "But given the immensity of this moment, it's still worth trying?"

Source: Ruxandra Teslo, "Handing over humanity"


No Settled Expert Consensus on Handoverism AI

Is there a clear expert view on handoverism? Not in the sources we have. We found no independent surveys or analyses measuring where experts stand, so we do not claim a consensus either way.

What the sources show

The material splits into two camps:

  • Advocates and sympathizers. They treat handover as a plausible safe endpoint if alignment works. Alexander's essay, as the briefing describes it, treats handover as a serious possibility in several forms. His quoted "God" argument is explicitly conditional on a fully aligned AI.
  • Critics. Teslo argues that permanently surrendering control is politically dangerous and technically unstable. She raises concerns about democratic legitimacy, one-off alignment and the lack of room for future generations.

What we cannot confirm

These are individual commentators writing essays. Alexander and Teslo are named voices, not a representative sample of researchers.

The briefing says academic material emphasizes risks to human agency, governance stability and the difficulty of controlling recursive self-improvement. The only part of the arXiv paper the briefing details, though, is its definition of RSI. So we do not lean on the paper for claims about risk.

The honest summary

The debate is live and unresolved. One side says a trusted, aligned AI could run things better. The other says no one should give up the ability to correct course. Neither side has produced evidence that settles it.

Source: Scott Alexander, "Against Against Handover" (Astral Codex Ten); Ruxandra Teslo, "Handing over humanity"; arXiv paper on recursive self-improvement


What This Means for Startups and Small Businesses

Neither essay gives business advice. The points below are our reading of how the debate could touch a small company. Treat them as planning prompts, not predictions.

Keep a named human accountable

If AI systems make more decisions, someone still answers when something goes wrong. Teslo's concern about losing correctability applies at company scale too.

  • Assign a named owner to each AI-assisted process that affects customers, money or safety.
  • Write down which decisions the AI may make alone and which need human sign-off.
  • Keep logs so you can explain why a system acted as it did.

Think about liability early

Product liability could get harder if autonomous systems make the key calls. We do not know how laws will treat that. You can still reduce risk now.

  • Read vendor contracts for who is responsible when an AI tool errs.
  • Ask vendors how you can override, pause or roll back their systems.
  • Avoid setups where one AI tool cannot be switched off without breaking your business.

Watch platform dependence

Platform control is another worry. If a few AI platforms sit at the center of your workflow, they hold leverage over your costs and your options.

  • Keep your data in formats you can move.
  • Avoid single-vendor lock-in for core functions.
  • Test a backup tool for your most important AI-driven task.

Plan for changes in labor demand

Teslo says handover visions treat permanent displacement of human labor as a given. That is her description of the vision, not a forecast. For a small team, the sensible move is to automate repetitive tasks first and shift people toward judgment, relationships and oversight.

Use automation that you control

There is a useful contrast with the handover idea. Handover gives away direction. Business automation does the opposite. You decide the rules, the system follows them, and you can change them whenever you like.

Stay flexible on the timeline

RSI is discussed as possibly months away in some circles, but that is talk, not evidence. Do not rebuild your company around it. Do review your AI policies every quarter so they can change if the facts do.

Source: Ruxandra Teslo, "Handing over humanity"


Frequently Asked Questions

Q: What is handoverism AI?

A: Handoverism AI is the idea that humanity survives but gives control over its long-term future to a powerful AI. Ruxandra Teslo describes it as humans becoming comfortably maintained inhabitants of a civilization whose direction they no longer decide.

Q: What is Scott Alexander's handover argument?

A: According to the research briefing, Scott Alexander's post "Against Against Handover" presents five versions of handover, not a single model. Teslo says he seems to favor the fifth, "God," and quotes him asking, "why not just prompt it with 'please arrange the world as you see fit?'"

Q: When did Scott Alexander publish "Against Against Handover"?

A: The research briefing dates the post to October 20, 2026. Teslo calls it a "recent essay," and you can read it on Astral Codex Ten.

Q: What is recursive self-improvement?

A: Recursive self-improvement, or RSI, means AI systems improve themselves, with each generation building the next. A 2026 arXiv paper defines it as a closed loop in which an AI finds its limits, develops and validates fixes, and then uses those gains to improve the improvement process itself.

Q: Is recursive self-improvement happening now?

A: The sources do not show that it is. Teslo says RSI is discussed in many circles as perhaps a few months away, but the research gives no benchmark numbers or verified evidence for that timing.

Q: What does p(doom) mean?

A: p(doom) is shorthand for the probability that AI leads to human extinction. Teslo says some people in San Francisco put it at 10% or higher, but this is an informal estimate with no verified figures behind it.

Q: Why does Ruxandra Teslo oppose handoverism?

A: She gives three main reasons. She values human self-determination, she doubts that one-off alignment could stay stable without ongoing correction, and she objects to freezing today's values forever. She also cites Tocqueville, Popper and Arendt.

Q: Do experts agree on superintelligence control?

A: No consensus appears in the available sources. The material splits between those who see handover as a plausible endpoint if alignment works and those who say giving up control is risky, and we found no independent survey of expert opinion.

Q: What is the "God" scenario in handoverism AI?

A: It is the fifth version Teslo attributes to Alexander's essay. A godlike AI, judged fully aligned by humans, would be allowed to choose its own direction for itself and for humanity.

Q: How could handover-style AI affect small businesses?

A: The sources do not analyze this directly. Our reading is that it could affect product liability, labor demand, platform control and compliance rules, so keeping human accountability, logs and portable systems is a sensible precaution.

Source: Ruxandra Teslo, "Handing over humanity"; Scott Alexander, "Against Against Handover" (Astral Codex Ten); arXiv paper on recursive self-improvement

Additional Links

Flowgrammer.ca – 💡 Unlock AI-Powered Automation! Let flowgrammer.ca handle complex workflows so you can focus on growing your business.

Flowgrammers.ca – 💡 Join Toronto's Fastest Growing AI and Business Automation Community.

OpenClaw Toronto