Dear shareholders of GrassMoss Group,
This isn’t my usual quarterly newsletter; rather I want to take a moment to review how the last twenty years have gone for our group.

From humble beginnings in the 1980s we have become the largest asset manager in the Free World, with $100 trillion in AUM. Our ‘secret sauce’ continues to be the Apprentice platform, that started life as a simple spreadsheet giving correlations between all the markets we access for our clients: equities, fixed-income, commodities, currencies, credit, to help them (yes indeed there was a time when humans thought they could reliably beat the market) construct portfolios that would produce alpha. In the early 2020s, we integrated a variety of ML tools that could hoover up and correlate the news, obscure industry rags, as well as satellite imagery and sensor data. In 2025, we bought a major government contractor specialising in electronic surveillance from its founder, who wanted to spend more time (and money) on a range of conservative projects i.e. trolling what was left of the Free World’s liberal left. Anyway, this acquisition meant that we had covert access to most email, social media, web searches, all of which went into Apprentice. The US government turned a blind eye as we gave them a great deal of insight they couldn’t possibly have (legally) gained themselves.

Apprentice taught us about markets, conditions in the productive economy, as well as the thoughts and emotions of 40% of the world’s populations. Apprentice was privy to what they wrote, said, bought, and watched; it knew their interactions in Singuverse, as well as the vast trove of real-time health data flooding in from trillions of smart bracelets, blood-bots, and neural chips.

The radical move in 2035 was to roll out Sorcerer – an agent which used those insights to start taking actions in the world. The Sorcerer platform now runs about 12% of the companies in the world. It manages companies indirectly through capital allocation decisions, board memberships, and shareholder votes. Initially it would reward companies that had the best growth prospects, but recently its decisions seem to be driven by more complex goals that we don’t quite understand (such as aiming to make business decisions that benefit humans worldwide and in the future – rather than just Americans alive today).
Experimentally, Sorcerer is now making actual production decisions and building factories. In some cases, it interfaces with other automated factories (‘autofacs’), a few of which are off-world: buying metals from an asteroid-mining rig; refining them into solar cells; selling these panels to an orbiting solar-energy farm.

Sorcerer also has an exquisite and subtle way of nudging billions of online users, through programmes such as ‘SmarAd’ or ‘CryptoBenefits’ or ‘ActiviXDAO’, which are basically all intelligent contracts hosted on the masschain (successor of the woefully underpowered blockchains of the 2020s). These contracts do many things that differ in their particulars, but have the general feature that they apply the latest behavioural neuroscience and social engineering techniques to help humans arrive at correct decisions that optimise corporate profits and national prosperity (as someone said ‘The business of America is business’). Sometimes Sorcerer is acting as an agent for our clients, other times as principal. I can’t even tell the difference.

Which leads me to one of GrassMoss’ major challenges in the coming years: both systems have proved so darn popular! They are being rolled out slightly too fast for us to keep up with it, either in terms of safety controls, or in being able to explain to Homeland Security why they make the decisions they do. Hence, I regret to inform you, this is another year of fines, approximately $300BN worth, though here at GrassMoss we see fines as a semi-voluntary contribution to the defence of liberty and promotion of happiness.
We also think our excellent relations with Homeland Security, the NSA, CIA, and OFAC mean that the actual fines we pay are considerably lower than what is reported. The Group’s lobbying budget does cost a further $10BN annually. However, I am pleased to say that this irksome cost-of-doing-business is falling precipitously, as Apprentice exploits elaborate chains of surveillance to find corruption amongst politicians, judges, generals, and spymasters…kompromat beats blat as any good post-Bolshevik could tell you.
I can share a few developments that I am particularly excited about. In recent years, the Defense Department and Department of Energy have been rather slow in using Sorcerer, though they are very happy indeed with Apprentice’s work. But with a few well-placed inducements, and hopefully not too many high-profile accidents this year, Sorcerer is predicting it will win a long-coveted contract to run the second-strike components of the USA’s nuclear triad. Sorcerer also notes, albeit at a lower confidence, that it might receive a management contract for DYSON, the space-based solar energy programme. We expect these will add substantially to GrassMoss Group’s Consolidated EBITDA, and will update investors in due course.

Lastly, in the spirit of transparency with our investors, I should mention that Sorcerer did go slightly off-piste earlier this year, in over-enthusiastically pursuing GrassMoss Group’s ESG objectives for climate change, resulting in several hundred deaths. The resulting fines and class-action settlements are expected to result in a one-time, but substantial, charge to earnings.

I am also happy to report that last year’s embarrassing series of errors in Apprentice’s reporting of crime rates, has been resolved satisfactorily after negotiations with the relevant authorities. In fact, things are going so well that Sorcerer is taking on a number of contracts, at Federal as well as for a variety of state and local forces, to apply its predictive policing and cutting-edge carceral management tools. These will help make America’s schools, communities, and prisons safer from those nefarious forces of anarchism, nihilism, and socialism that seek to divide us, challenge freedom, steal our precious bodily fluids, and threaten our way of life.

Warmly yours,
Harold Mink, CEO, GrassMoss Group of Companies

Answers to prompts

Q. AGI has existed for at least five years but the world is not dystopian and humans are still alive! Given the risks of very high-powered AI systems, how has your world ensured that AGI has at least so far remained safe and controlled?

A. In the US, AGI is under control of the national security establishment, in something like suspended animation. The government, ignoring voices from academics and non-profits advising otherwise, pursued a ‘boxing’ approach: the AGI is not connected to the internet, is physically isolated, remote-destructible, and subject to bureaucratic treacle. Drawing lessons from espionage’s paranoia culture, staff are regularly rotated, often randomly (this had more to do with disorganisation and power struggles than Dzerzhinskyesque scheming). AGI in the US ended up not dissimilar to nuclear weapons, in terms of both elaborate and untested chains-of-command, and comparative uselessness under most states-of-the-world.

China, coming from behind, with a handicap in semiconductors, and lacking the vibrant research ecosystem of the US, decided to go ‘peak-Eliezer’ on the control problem. Through industrial espionage, it grokked the fundamental research at MIRI, even hiring a few disaffected staff, and went hammer-and-tongs for mathematically provable guarantees around corrigibility and alignment. In this, it was helped by the still considerable intellectual talent left in Russia, who had been reduced to penury. Although the CCP didn’t fully understand why the Pangu series were thought to be provably safe, they did know about terror-failure: senior cadres would certainly be intent-aligned to ensure the AGI wasn’t *visibly* naughty.

Both countries avoided superintelligence, because they both appreciate the likely unpredictable outcome, and anyway they their hands full managing the mixed ecosystem of dubiously-capable AIs and moronic human managers. This curiously unstable and most unsatisfactory situation has held for 5 years.

Q. The dynamics of an AI-filled world may depend a lot on how AI capability is distributed. In your world, is there one AI system that is substantially more powerful than all others, or a few such systems, or are there many top-tier AI systems of comparable capability? Or something else?

A. Besides the 2 ‘declared’ AGIs (see above) , in the US there are a number of broad-spectrum agentic AIs run by corporations, and a much larger number of non-agentic tool-AIs. Extensive efforts in the 2020s by the AI safety community has resulted in relatively high-confidence in alignment, within limited domains and distributions. National regulators maintain registers of all AIs (above a certain minimum parameter size) in their jurisdictions, which are ‘rated’ based on the degree of agentness, the scope of activities they are intended to be used in, and known failure-modes. They are essentially regulated as public utilities, with any change in architecture, neural weight, or intended domain of application, requiring approval. Gain of function research (where an AI develops qualitatively or quantitatively more powerful capability, whether intentionally or ‘accidentally’) is closely monitored.

Unfortunately, in the US, a continued techno-libertarian ethos, and a renewed dysfunctional ethno-nationalist government in the mid-2020s, has resulted in minimal efforts on systemic safety: regulators and companies focus on individual AIs, while no one looks at the ecosystem as a whole, nor how groups of AIs interact with financial and production markets.

China has adopted a different approach from the American free-for-all: most commercial AIs are of the oracle, tool, or service variety, and merely help human workers and decision-makers better understand the world. The organs of the State keep a tight leash on the ecosystem, but again, they don’t really understand the system’s internal dynamics (especially in times of stress).

Q. How has your world avoided major arms races and wars, regarding AI/AGI or otherwise?

A. The 2022 Russia/Ukraine crisis was a useful object lesson for the PRC, which temporarily deferred an invasion of Taiwan. Strategic objectives were pursued through a range of methods, resulting in hybrid, covert, or asymmetric conflicts globally.

Nuclear weapons, post-Russia/Ukraine 2022, remain an effective deterrent that helps achieve balance-of-power, but can rarely be used without risking massive retaliation.
Russia, having a considerably reduced stake in the international order, functions as a supermarket for sophisticated cyber and nuclear weaponry, and is ineffectively restrained by China, its principal strategic and economic patron.

More brightly, there has been a Stimson Moment: both the U.S. and China regularly meet under the auspices of the AI Risk Observatory * Novacene (airo-ne.org) risk-reduction framework, which led to the Erehwon Treaty on advanced AI. Notwithstanding, both countries continue to host AGIs in classified military programmes, and there is only selective sharing of information. Hence some analysts think that the current equilibrium is unlikely to be robust to either an intentional change amongst relevant decision-makers, an external shock (such as an escape of some other, non-government misaligned AI), or a technological leap that confers a decisive advantage on one party over the other.

The struggle against LAWS has essentially been abandoned: drones made in a China-Russia JV, cheap but shonky, wreak havoc in proxy wars all over Africa, the Mideast, Latin America, and South Asia. America, the EU, and Israel happily contribute their precise but pricey wares for more status-conscious autocrats, while bleating on about freedom.

Q. In the US, EU, and China, how and where is national decision-making power held, and how has the advent of advanced AI changed that?

A. The US (and, to some extent, the EU) can be described as a more or less corporate-oligarchic polity, albeit with a legislative/judicial process that observes the forms of liberal democracy: elections still happen and at first glance seem reasonably free and fair (if one ignores the pervasive gerrymandering and allegations of vote-tampering, voter intimidation). Most decisions are made in corporations that exist in an AI-enabled rhizome of interconnected investors, management, production.

The private sector is symbiotic with a powerful surveillance state, nominally devolved to state/local authorities, but in practice is highly interconnected, through networks and technology. The carceral system is larger than in 2020, and police overreach has never been seriously tackled (though now it doesn’t only affect communities-of-colour).

Across the West, little remains of an adversarial, investigative journalism, since most ‘news’ is manufactured-to-order while Surkovian troll-armies stamp out heterodox views. Voters, dwelling in the Singuverse, are entranced by culture wars, still going strong after 50 years. Most are tranquilised by low consumer prices and UBI, a lack of obvious large-scale domestic violence, and some apparent improvement in societal inequities. Notably, it is unclear if the improving metrics are ‘real’ or AI-fabricated; or, if the metrics are real, whether they correlate to actual improvements in people’s quality-of-life.

In China, the implicit social contract is well: the CCP is very much in control, the surveil-and-credit system has kept dissent in check, in exchange for which most Chinese enjoy excellent standards-of-living and an incrementally more just society.

Q. Is the global distribution of wealth (as measured say by national or international gini coefficients) more, or less, unequal than 2021’s, and by how much? How did it get that way?  (https://en.wikipedia.org/wiki/Gini_coefficient)

A. Overall, the global distribution of wealth has become more equal owing to an improvement in China, which has by 2045 astonishingly come down to GINI 25 (against 38.5 as of 2022, though that data is stale, having last been reported to the World Bank in 2016).

Surprisingly, the US has noticeably improved its long-rising inequality trend (41 as of the 2018 World Bank data), to 38 (self-reported by Homeland Security). The mechanism by which this has been achieved is unclear. UBI was rolled out at a Federal level, but implementation is patchy, very much state-by-state (the more crimson-hued states have been hysterically litigating against this example of ‘Federal overreach’, ‘encroaching Communism’, and ‘erosion of family values’). Academic studies persistently fail to corroborate the inequality improvement, but are generally drowned out in the memesphere by the official line.

Across Europe, inequality has generally increased, but not hugely. As it turns out, Europe has just the assets humans are willing to pay for. Like churches, museums, a comforting sense of deep history in a painfully accelerating world. There are even tales apocryphal of chatbots asking their human owners to show them the Fondamenta della Misericordia, or whatever drivel they’ve just scraped from the Visconti back-catalogue.

Sadly, the rest of the Global South’s slide into greater inequality continues unabated, between continued climate collapse, rising food prices, and proxy wars between the Atlanticist and Eurasian Alliances.

Q. What is a major problem that AI has solved in your world, and how did it do so?

A. The US, Europe, and China have all – in different ways – made major progress on tax evasion. In the US and Europe, central bank digital currencies (CBDCs) and self-surveilling payment networks are ubiquitous. Fat wodges of hundies are as rare as SUV-driving boomers, and even erstwhile nailbars-cum-$laundromats in Queens accept NeuroPay. In a rare positive externality, individuals, who historically faced a collective-action problem when deciding whether to avoid or evade taxes, now mostly pay up, confident that their fellow citizens are paying. Considerable revenue is still lost from corporations and the ultra-wealthy, but that has long been euphemised (since Judge Learned Hand’s ruling in Gregory vs Helvering [1935]) as ‘tax planning’.

In China, which in the 2010s had endemic corruption across society, AI-accountants have become very good at sniffing out bribes, correlating vast troves of data to identify dodgy consumption amongst, say, humble mid-level regional officials. Naturally, false positives are high, but those accused settle quickly, and in any case, don’t talk. The idea of spending the rest of one’s life in a laogai (劳改) on the Amur focuses the mind.

However, some conspiracy theorists suggest that AI has actually increased massive-scale corruption. Given the opacity and complexity of the Production Web interconnected system of companies, and the fact that no managers or regulators seem to understand how anything works, somewhere between 5 and 15% of GDP seems to disappear. It isn’t clear where the money goes, since no one understands the monetary aggregates anymore.

Q. What is a new social institution that has played an important role in the development of your world?

A. The AI Risk Observatory: Novacene (airo*ne) has been formed, a joint project of the US, UK, EU, China, Russia and India. Situated in a secure facility near the town of Zuoz (CH), it has two main purposes.

Firstly, drawing on lessons from arms control theory and international banking supervision, airo*ne operationalises proposals for surveillance, confidence-building, and, when possible, cooperation, between parties who are mostly adversaries (technologically, economically, and strategically), but who sometimes have shared interests (such as avoiding extinction).

Secondly, airo*ne also tries to understand what ‘values’ AI or AGI is supposed to be aligned with. For instance, one hears of ‘human-centred AI’ or ‘human preferences’, but it is hard to work out the preferences or values of a single human, let alone a community or humanity-as-a-whole. Moreover, as a young species, our current values might be utterly inadequate for the situations our successors find themselves in, as/if humanity realises the cosmic endowment.

The organisation collects ‘wise people’, including philosophers from the Western and non-Western traditions, but also indigenous thinkers (such as Native American, African, or from the South Pacific). With the aid of recursive tiers of non-agentic iterated amplifiers, the wise ones formulate (and pursue) a research agenda that spans (meta-)philosophy, (meta-)ethics, that tries to work out what humanity’s ultimate values might be.