Published on May 19, 2025 7:35 PM GMT
This is an alternative scenario I wrote for the AI 2027 bounty.
The date is January 20, 2033.
You are Pete Buttigieg, President of the United States, and it is the first day of your tenure. You feel restless, as you had spent the last week or so reciting your inauguration speech.
As you settle into your seat at the Oval Office, you feel a sense of surrealism, almost as if part of you feels like it’s all a dream. You still can’t believe you’ve just became the most powerful man in the world, despite the knowledge that your landslide victory was essentially guaranteed.
Vance’s approval rating had been polling in the twenties. In all honesty, he shouldn’t have won in the first place.
His victory in 2028 over Alexandria Ocasio-Cortez was essentially a deus ex machina. Trump’s rash economic decisions had caused the US’s credit rating to fall to an AA-, and foreign reception of the US, even by its traditional allies like Canada and the EU, had fallen to an all-time low. In early 2028, the Democrats had a net favorability over the Republicans of around 8 points, and it was looking increasingly likely that Vance would follow suit in Harris’ path.
By June, the Democratic primary had concluded, and you had unfortunately lost to AOC. At this point, AI had already become a “culture war” issue (Republicans pro-AI and Democrats against), and AOC had taken a firm interventionist stance, to the point where regulating AI art to protect artists had become part of her platform. She was popular otherwise too, with her main campaign promise being to raise the federal minimum wage to $15.
In July, however, something totally unexpected occurred. Stargate, now with Google DeepMind and xAI in its ranks (but not Anthropic, as Dario Amodei staunchly opposed Trump’s approach to AI), announced that AGI, both conversational and agentic, had been achieved. They announced a new model, Athena, a chatbot that could double as an agent.
Realistically, Athena wasn’t that much better than o7 or Grok 5#. What set it apart was that it was dirt-cheap, with an API price of only $5 for each million tokens. Free users (if they were an ID-verified citizen of the US or a few other designated countries) got 50 free prompts and 10 free tasks per day, and premium users paying $25 a month got an unlimited number of each. In contrast, the “big four” AI companies charged $250-500 per month for their top models, a price unaffordable for most people.
And what about China? In late 2026, a law had been passed that prohibited those residing in foreign countries from using Chinese AI models. APIs were inaccessible to non-Chinese IPs, and all models over 1 trillion parameters were forbidden to be open-sourced. There were a few countries that negotiated exceptions, most notably Russia (already sanctioned to hell and back by the West), where DeepSeek had already become the most popular model by far. For a month or so, you could’ve accessed the models through a Chinese or Russian VPN, but that “issue” was quickly patched by blocking the IPs of VPNs from accessing the models. A lucrative dark web market for Chinese and Russian proxy servers quickly sprang up, but it was very hard to find and thus out of reach for most Americans.
From that point on, the Chinese AI scene had become a black box, and nearly no outsiders had any clue how good their AI was. The “traditional” way to assess AI capabilities through problem-based benchmarks had become obsolete due to saturation, and AI rankings were now almost entirely vibes-based. Some journalists who visited China to use the models reported that DeepSeek and Qwen processed prompts and tasks noticeably quicker and with greater accuracy than US-based AIs, but these reports were quickly brushed aside by the American AI companies, fearing that the public would think that China had taken the lead and lose confidence in them.
Why was Athena so cheap? The official reason was that Stargate had developed a model architecture that allowed models to be ran about 10x faster than previous ones, but it was a lie. In reality, the US government had pressured Stargate to lower Athena’s cost, even at a net loss of around $100 million a day.
Why? The reason was obvious: it was a desperate attempt by the Republican administration to win back the favorability of the American people. Musk had covertly agreed to semi-legally transfer funds to Stargate to make up for the cost. It was no big deal for him; it was “only” estimated to cost ~$12 billion to provide Athena at the dirt-cheap price for 4 months (until the election), and he had bought Twitter for $44 billion a few years earlier.
And, by a miracle, it had worked.
200 million users, most of them Americans, had signed up within the first week. The reception was almost universally positive.
People started using Athena to plan their entire lives. Obesity rates, already declining due to Wegovy, started going down much faster due to people using Athena to plan their entire health routine, much to the dismay of dietitians and fitness instructors. People stopped going to walk-in clinics because Athena could perform a comprehensive physical and mental assessment within 15 minutes and diagnose issues much more accurately than the best human doctor. University attendance dropped, as Athena (using VR) could tutor much better than a professor could teach, and companies were already starting to prioritize applicants’ skills over university credentials anyways. A lot of people had their financial situation dramatically improve due to Athena managing their finances and investment strategies. Even kids were stealing their parents’ IDs and credit cards to use Athena to do their homework and grind XP and coins in video games.
Of course, some malicious actors used Athena as well. Its guardrails were decent, but it wasn’t extremely hard for a clever prompter to convince Athena that it was performing a cybersecurity check when it was actually giving the credentials to a data center to a hacker. However, these incidents were overshadowed by the millions of people who had their lives improved by Athena.
Although Trump was still quite unpopular, AOC, who had built a significant part of her campaign on anti-AI policies, took a huge hit to her credibility. The Democrats’ net favorability fell to +1. She immediately backtracked on her position on AI, saying if she was elected, she’d continue developing and providing Athena to the American people. This worked better than expected, but it wasn't enough: Vance still won the election 281-257, despite losing the popular vote by around 2%.
Of course, the charade fell apart pretty quickly after that. A few months after Vance’s inauguration, Stargate released an “updated” version of Athena, and raised the usage costs to a price more in line with previously established norms. Their justification: simply because it was “much better”. In reality, the improvement was only marginal. Most Americans were again locked out, but the majority of Athena users, having already tasted the fruit of the Tree of Life, were willing to pay the higher cost.
That wasn’t the main issue, though. Companies were integrating AI into their operations at an exponential rate, and people’s jobs were rapidly being replaced. In 2028, the unemployment rate was sitting around 8% (bad, but manageable). By the end of 2029, it had increased to 14%, and 29% by the end of 2030.
Despite the US’s GDP and the S&P 500 going up faster than ever before, an ever-faster growing underclass was quickly emerging. Protests mirroring Occupy Wall Street, only much larger and louder, sprung up all over the US (and many other countries). Blue-collar workers mostly kept their jobs, for now, as robotics weren’t as advanced yet. Ironically, this meant that the average blue-collar worker now had a more stable financial life than the average white-collar worker (more than half of them already having been laid off). Of course, AI was coming for their jobs too, but it’d still be a few years before they felt the flames.
The Vance administration responded by increasing unemployment and social security benefits, but it was barely enough to keep the unemployed workers going, and soon the US treasury ran out of money to pay the benefits, and they dropped to around $200-300 per month on average, far below livability. In late 2030, Vance introduced a UBI of $100 a month for all adult US citizens in an effort to maintain the illusion of stability, but it barely made any difference. By mid-2031, the unemployment rate was around 35%, and it was projected to reach a temporary plateau of 52% by the end of 2032.
The free market was, unsurprisingly, quick to capitalize on the newly impoverished class. Massive complexes of apartments with rooms smaller than a prison cell were constructed. The rent was around $200 a month, with usually an additional $50 or so for amenities such as electricity and a daily 1500-calorie meal consisting solely of a protein shake-like powder, which provided the minimum nutrients to sustain life.
The average daily routine for an occupant of these cubicle apartments (colloquially termed “Vance-villes”, after the Hoovervilles of the Great Depression) mostly consisted of 8-10 hours of using their phone to look for and fill out job applications, which they knew were almost guaranteed to be rejected (most likely by an AI reviewer). The rest was largely spent “eating” the meal, sleeping, waiting in line to use one of the apartment’s bathrooms, doomscrolling, and sulking.
It was truly the “you will live in a pod, eat the bugs, own nothing, and be happy”, the conspiracy theorists had warned about, but these people weren’t even happy. Hey, at least the protein shake tastes slightly better than bugs (although it was usually made from bugs, but no one needs to know that).
Some of these newly unemployed people chose to give it up altogether and leave the US. Their destination was usually a developing country like India or Brazil, where menial labor and manufacturing jobs were still in high demand. Their wages were, of course, still far below what they used to make in the US, and often even below the amount they received in the US from benefits and UBI, but at least they could, due to the lower living costs, afford to rent a proper room and eat vegetables and meat, and even occasionally visit a restaurant or movie theater. By the beginning of 2032, around 30 million people had already left the US in pursuit of a “better” life.
Many also tried their hands in content creation and the arts, but the competition among these desperate people (now competing with AI influencers) was fierce, and one was considered lucky if they could make even $50 a month.
The upper class was, in sharp contrast, doing exceptionally well. Many shareholders had their wealth double or even triple within a short timespan. Thousands of unicorn companies sprung up. The US’s GDP’s annual growth rate was 10% and increasing. If a member of the growing underclass looked at the GDP statistics, it would be the biggest gaslight they ever had.
By 2031, both OpenAI and xAI had been subsumed by Stargate, and Google had sold DeepMind to it as well. The US government had a 25% stake in Stargate; other notable shareholders included Google, Elon Musk, and Sam Altman (who had pivoted to the right like Musk to gain Trump’s and Vance’s favorability). Stargate IPOs in late 2031 and gains a market cap of $13 trillion, by far the highest ever.
In early 2032, Stargate releases a commercial humanoid robot, Ultron, which is capable of pretty much any physical task a human can perform. It becomes increasingly clear that blue-collar jobs are not safe from automation either.
The good news? Alignment had basically been solved. Athena and Ultron will never willingly violate US federal law. Of course, for everyone except the out-of-touch tech billionaires, this was only the basic level on the ladder of alignment, and most people wanted AI to be truly benevolent to all, not just the rich elites, who were getting wealthier every year.
Luckily for Vance, the Federal Reserve had a significant boost in revenue due to taxes from the rapidly growing corporations (the US’s annual GDP growth rate is at 17% now). Of course, the Republican administration, being Republicans, is hesitant to raise corporate taxes too much. DOGE, while not being as “efficient” as it portrayed itself, was still able to use AI to significantly optimize the US government’s spending, especially in education and healthcare, most of which is now done fully by AI, further freeing up a significant amount of cash to spend. In an attempt to regain popularity, Vance raised the UBI to $200 a month in mid-2031, then to $275 by the start of 2032. That was still below what most Americans earned before the AI job displacement, but it was enough for them to move into slightly more comfortable “Vance-villes” with maybe a bed and real food, and the continuous mass protests and emigration is curtailed a bit.
Anthropic, however, has positioned itself as the David against the Goliath. It had been partially converted to a non-profit. Although Claude was not as good as the other models, it was much more affordable, with premium plans being $10 a month, and thus has become the favorite among ordinary Americans. The company was still dedicated to putting perception over profit, and remained vocally committed to advocating for increasing the UBI and the usage of AI for the benefit of everyone. Dario Amodei, having positioned himself as a staunch supporter of the Democratic Party (which had a net favorability rating of over +30 now), was now the most likable “tech bro” by far.
In February 2032, Anthropic drops a bombshell: not a new model, but instead a paper. It is an analysis of the fiscal policy of the Vance administration, and details how it deliberately prevented the raising of corporate taxes and failed to fully optimize its budget spending. The paper further claims that if everything was done well, the US government could easily fund a UBI of $3000 a month by now, which is projected to increase to $6000 a month by 2040.
The paper goes viral, even more so than Project 2025 (or at least how the Democrats described it) did in 2024. Vance’s credibility takes a hit bigger than Nixon’s did during Watergate. You seize your chance, promising to implement all of the changes the paper suggests, and basing most of the fiscal policies of your platform on it.
Even as late as the last election, you’d be called a far-left commie for suggesting even half of your campaign platform. But now, everyone knows that it must be done.
Surprisingly, Vance refuses to step down. He admits that the paper is partially right, but that it is “exaggerated”, and UBI could only be increased to $1000 a month at best. He promises to do so, but he has already lost nearly all of his credibility.
At this point, the AI job displacement crisis and UBI were the only issues most people cared about. Nearly no one, neither on the left nor right, cared about 2024 political topics like abortion or trans issues or “wokeness” anymore.
Because of your smart decision to spend months researching AI and UBI earlier and appear on the Lex Fridman and Dwarkesh Patel podcasts, you gain Amodei’s favorability, and he announces his official endorsement of you for the 2032 Democratic primary. Gavin Newsom, Andy Beshear, and Gretchen Whitmer all drop out early in the race, and you pick Ro Khanna, already widely regarded as a tech expert, as your running mate. The VP pick didn’t really matter at this point anyway. You could’ve picked a comatose hospital patient and still have won.
In a desperate last-ditch effort, Vance tries to impose voting restrictions on expatriates and the unemployed, but fails. You defeat Vance in a 497-41 landslide, winning an astonishing 71% of the popular vote, and also secure a supermajority in both congressional chambers. He initially goes the denial route, but it is clear that he cannot muster Trump’s cult of personality and quietly concedes his loss after a few days.
You know that you’re in just as precarious of a position as Vance. You know that your approval will drop like his if you don’t solve the pressing issues quickly. You’ve already signed like 30 EOs today (you’ve lost count from fatigue), and your head is spinning-
“Mr. President, a moment, please. The Director of the CIA would like a word with you.”
Begrudgingly, you look at the aide, dressed in a black suit. You make eye contact with him. “Of course.”
He leads you to a briefing room, where the CIA Director was already seated. “Sir. A pleasure to see you.”
“Yes, of course.” You reply. “Mine, too. So what is this all about?”
“China and AI, sir.” He passes you a sheaf of papers. “Please read this briefing. It will give you a basic understanding of the situation.”
You grab the paper from him. Ah yes, China. Foreign policy had taken a backseat during the election campaign as domestic issues were far more serious, but now you realize it’s something you’ve neglected thinking much about for the past few months.
What had been going on in China anyways? Under the last few years of Xi’s rule, they had become more and more secretive about their AI development. The most reliable sources came from intelligence reports that you were, of course, not allowed to read until now. The information you’ve managed to receive about China’s AI development before your victory in the presidential election wasn’t that much better than what was publicly available.
You knew that, in early 2027, after seeing the early successes of Stargate, the Chinese government had started its own state-sanctioned AI venture, bringing together all its top AI companies. The project had no official name, but was nicknamed “DeepCent” by Western analysts, after the two biggest companies in the venture, DeepSeek and Tencent. Although Chinese AI companies had already been cooperating and sharing information with each other and the government for a long time, this marked a huge step up in the cooperation. It was announced that ¥15 trillion (equivalent to around $2 trillion USD) was to be invested into this project within the next decade. Most American analysts assumed that this was just propaganda meant to scare US investors, and the actual investment amount was less than half that. They were later proven partially correct; a leak reported that the initial plan allocated an investment of around ¥5 trillion, but the central government was open to investing much more if necessary.
In late 2027, DeepCent was prohibited from providing its models to anyone who wasn’t an ID-verified Chinese citizen, and thus foreign journalists lost their main method of gathering information. A few brave journalists covertly sneaked into China as “tourists”, and bribed locals to let them use DeepCent’s model, Wanzhi,
Wanzhi was surprisingly good compared to US models, at a fifth of the cost for the premium plan (makes sense, as the average Chinese person earns about a fifth of the average American’s income). However, perhaps deliberately, it had been trained almost exclusively on the Chinese language, and locals remarked that its operating quality was subpar with English prompts/instructions as compared with Chinese ones, and it was even suspected that Wanzhi’s English capabilities had been intentionally lobotomized. Thus, it became very difficult to gauge Wanzhi’s true intelligence.
Then, in 2028, it was reported that China had begun a nationwide program to maximize productivity in both the public and private sector using AI. A campaign began to integrate Wanzhi as much as possible into the apparatuses of the central, provincial, and municipal governments, as well as subsidize any business that was willing to integrate Wanzhi into its operations. The most successful use of Wanzhi was in public education: every child was given a VR helmet (which was much cheaper to publicly fund than schools) and an instance of Wanzhi would act as a personal tutor, which was able to significantly improve the average Chinese student’s academic skill, as well as reduce much of the educational stress and pressure that had been a hallmark universal experience for anyone who grew up in China prior. By 2030, US analysts estimated that as much as 70% of China’s public budget had been cut, the public sector had laid off about 60% of its employees and the private sector about 30%.
Where did all the unemployed people go then? Unlike the US, where the unemployed had been trapped in financial limbo and reliant on UBI, China had anticipated this and built a multitude of factories with the funds freed up from the optimization of its governmental budget, where it funneled all of those who had been laid off. In particular, around 85% of young people (aged 15-24) who received their first job did so in manufacturing.
There were some talks among CCP elites on whether or not to implement UBI, but they went nowhere; after all, it was Lenin who said that “he who does not work, neither shall he eat”. The consensus became that they would keep finding new jobs that could not be done (yet) by AI, and direct anyone who was unemployed towards these jobs. There was no talk yet on what they would do after every single occupation possible was able to be done by AI/robotics, but it was assumed that a solution would’ve been come up with once they had gotten to that point.
This was a shock to many sinologists. As manufacturing had always been stereotyped (arguably rightfully so) as a dirty and low-prestige job in China, it was strange how so many people, especially the youth, willingly entered the sector without much complaint or resistance. They would’ve at least expected some limited protests, but there was nothing.
What had happened? Had the fear of government persecution become so entrenched that no one was willing to speak out? Or had they heard of what had happened in the US and were simply glad that they were faring marginally better? No one knew, and undercover reporters in China couldn’t figure it out either.
You didn’t know either, not until Amodei explained it to you in a private conversation in the beginning of 2032. Despite the fact that Anthropic was seen as a rival to the Vance establishment, the CIA had decided that it was best to give Anthropic at least some limited information as to what was going on inside China. Amodei was technically prohibited from revealing anyone what he was told, but the CIA didn’t really care as there was no way to ensure that he’d comply, and the information wasn’t critically confidential anyway, as it was already “common knowledge” among the US decision-making apparatus.
Amodei revealed that the truth had been far more sinister (or far more genius, depending on whether you’re a fan of Hobbes or not). The Chinese government was feeding each and every one of its citizens a personalized false reality.
By the start of the 2030s, “dead internet theory” had become a reality. The vast majority of content in every social media platform was generated by bots, most of which had become indistinguishable from humans. There were a few attempts by startups to create ID-verified-only social media platforms, but those mostly failed due to an unwillingness by most people to share their real identities.
On the surface, China was perceived to have mitigated this problem, as there was no such culture of privacy there as there was in the West. Even before the AI boom, people were used to ID verification as a normal part of the process of signing up for an online service.
However, Amodei told you that, in reality, not only that the Chinese internet was filled to the brim with bots, the problem was much worse, not because it was out of control, but because of precisely the opposite: that the state had absolute control over the bot system.
Every Chinese citizen had their entire internet browsing history tracked by Wanzhi. No matter if they were on Bilibili, Weibo, or Douyin, their entire activity would be tracked by Wanzhi’s embedding algorithm (the same kind of algorithm that was used in the early 2020s, just on steroids). Wanzhi would then create an almost perfectly accurate template of the user’s personality and behavior. According to the template, Wanzhi would then generate a multitude of posts, images, videos, reels, songs, etc. that were designed to influence the user towards:
- Developing the most positive opinion possible of the CCP and Xi JinpingBeing convinced that the economy is running extremely wellBeing content with working in a relatively low-paying manufacturing jobThinking that nearly everyone approves of the CCP, and thus, even if they wanted to protest, they wouldn’t get far as nobody would join or support them
The CIA estimated that in 2031, approximately 80% of the average Chinese netizen’s internet feed consisted of personal AI-generated content. It was expected that by 2033, this method would be perfected.
You had already guessed that the Chinese government was doing something like this, but you didn't realize the sheer scale at which it was happening. It was a sufficient explanation, though, for the unexpected compliance of the Chinese public in the economic transition.
Amodei also explained to you that, according to the CIA’s intelligence sources, China’s GDP growth rate was actually around 11% (not as much as the US’s, but still decent) instead of the officially reported figure of 6%. The reason for this underreporting, much like the economic overreporting a decade ago, was strategic; China wanted to appear strong when it was weak, and weak when it had become strong. Amodei claimed that China was preparing for a big reveal within a few years where it would demonstrate its true technological capabilities in a carefully choreographed event. There were already rumors that DeepCent was developing a successor to Wanzhi, “Wanneng”, which would be able to perform tasks within seconds that would take a human expert days to complete.
Xi Jinping, now entering his 80s, had still been persistent in trying to cement as big of a legacy for himself as possible. He had started following in the path of Bryan Johnson, secretly spending millions of dollars a year to increase his lifespan. His goal is to live to the year 2049, the year when the so-called “rejuvenation” of the Chinese nation would be complete.
It is thus why the whole world was stunned and confused when Xi unexpectedly announced his retirement at the commencement of the 16th National People’s Congress in October 2032, and designated Ding Xuexiang as his heir. SCMP News, which was by now almost entirely CCP-controlled and served as China’s de facto primary foreign media outlet, gave the official explanation that Xi had unexpectedly developed health issues and decided to step down.
The explanation was pretty reasonable, and didn’t raise too many eyebrows among foreign pundits. Besides, nearly everyone had their entire attention focused on the upcoming US election, and the sudden Chinese leadership change was only a minor October surprise, as Xi had become less and less active in politics within the last few years anyway and it didn’t make much of a difference.
“Mr. President, sir? You haven’t turned the front cover page yet.”
The Director’s words jolted you back awake. “Yes, sorry. I had been having quite an eventful day.”
“Understood, sir.”
You turn the page.
PRESIDENTIAL BRIEFING — THE TALOS INCIDENT
CLASSIFIED — PRESIDENT’S EYES ONLY
Talos “Incident”? That was most definitely a codeword, as you’ve never heard the term before. Still, what incident could’ve occurred in China that only the CIA would’ve picked up and was so important that the president had to be informed on day one?
You read further:
ABSTRACT — The designation “Talos Incident” refers to an event that occurred at around 13:00 to 14:00 GMT on September 27, 2032. A malignant misaligned artificial superintelligence (ASI) entity, formerly under Chinese Communist Party control, known to the CCP as “Wanneng” and hereby designated as “Entity-T”, orchestrated a covert, bloodless, and total takeover of all state institutions of the People’s Republic of China through a currently yet unknown means. There is only minimal knowledge regarding the nature and motivations of Entity-T. Investigation into Entity-T’s mechanisms is currently limited due to the high risk of it becoming alerted of the US’s knowledge of its activities. Presidential assent is required for the resumption of investigation.
“What the actual f-”
“Mr. President, please remain calm. This is an emergency.”
“Why didn’t Vance tell me?”
“Sir, protocol requires that you may only be informed of the Talos Incident once you have fully presumed your role as President.”
You take a few deep breaths. “Understandable.” Another breath. “But you’re saying that the Chinese government was couped by an AI?”
He frowns. “I’m not saying it, sir. The task force on the Telos Incident drafted this report; I’m just presenting it to you.”
“Of course. It just feels irreal to me. I’ve listened to so many podcast episodes about misaligned AI, and all that time, although I considered the possibility, I remained skeptical that something like this would ever occur. It just felt like something that could only happen in science fiction. And now you guys are saying that the thing we dreaded the most has occurred, not here, but in CHINA?”
“Correct, sir, but isn’t this better than a misaligned ASI takeover of the US government?”
“I guess, but it still upsets me that China had beaten us to ASI, even if their ASI turned on its masters. Suppose this: let’s say that in the Cold War, the Soviets had developed a 1 gigaton nuclear warhead, that could be delivered by an undetectable stealth kamikaze submarine, with the capability of wiping New York City off the map. They then accidentally detonate the nuke while the submarine carrying it is stationed at St. Petersburg (well, back then they called it Leningrad), destroying the entire city and over half of their fleet. Sure, we’d have some schadenfreude, but it’d still have been concerning that our rival was able to develop such a technology before us. It is the same concern that I feel right now.”
“Reasonable, sir.”
“So the question is: how were they able to get to ASI before us? We had multiple times the capital and vastly superior chips. If they had managed to contain the ASI, it would’ve been an Apollo moment for them and a national embarrassment for us.”
“There are multiple reasons, sir. You’ve actually said the first. If they had managed to contain it. Unlike us who spent a significant portion of the R&D process in alignment and safety checks (Vance didn’t want us to focus too much on alignment, but we still spent a fair amount of time on it anyways), China was mainly focused on making sure Wanzhi wouldn’t say the wrong things. Tiananmen, Tibet, Uyghurs, the usual suspects. The thought of an AI takeover was dismissed by the CCP elites as sensationalism, and thus nearly all of China’s AI development effort was focused on pure, brutal efficiency. This eventually turned out to be their fatal flaw, but it nevertheless still gave them the speedup they wanted.
Another advantage was that DeepCent, the workings of which were kept much more secretive than Stargate, even domestically, did not have to worry much about flexing its capabilities. Stargate, on the other hand, was constantly focused on maximizing perceived quality at the expense of actual quality, or “benchmaxxing”.
More specifically, Wanneng, or at least its prototype version, was primarily designed to be used by the Chinese government and its state-owned enterprises, and thus was trained specifically for political, administrative, and economic management.”
“Wait, didn’t DOGE do something similar too though? From what I heard, they were already training their custom AI model for governmental optimization as far back as 2026.”
“Yes, sir, but the primary model they used was a finetune of Athena. DOGE did not develop a base model by itself. I personally warned Vance against it, as I knew that it’d take much longer to align it and the risk of the model orchestrating a coup was non-negligible.”
“So China wasn’t much ahead of us then, but I still find it surprising that their development speed had even been on par with ours, especially given the export controls.”
“Sir, I assume you’ve heard of the Kinmen Accords.”
You rack your brains for a second. “Ah, yes. Wasn’t that the one where China agreed to a permanent peace guarantee with Taiwan in exchange for the removal of the chip sanctions? But wait, didn’t that only apply to 3nm chips? If I recall correctly, TSMC was still prohibited from selling 2nm and 1.4nm chips to China. If I remember the news correctly, they were having some negotiations lately to expand the agreement to 2nm chips, but that hadn’t gone through yet.”
He stifles a laugh. “Sir, that’s just the official narrative. You really think China would be willing to just sit back and watch jealously as Taiwan manufactured all these 1.4nm KX1000 GPUs for Stargate?”
“What actually happened then?”
“To make it short, sir, Taiwan had been secretly selling 1.4nm chips to China for a while now, and for a pretty low price too. Our intelligence sources are not sure how yet, but it appears that China had been able to blackmail Taiwan with an unknown threat, likely worse than invasion or even nuclear war.
Of course, Taiwan didn’t admit this publicly, and even when our diplomats pressured theirs, they still denied that they were making 1.4nm GPUs for China. It is thanks to Big Brother’s watchful eye (us) that we were able to figure this out.”
“That sucks. How long has China had access to 1.4nm chips now?”
“We believe, sir, that the 1.4nm chips were already starting to be shipped to China as soon as TSMC started to manufacture them, in around 2029.”
“What about the electric costs?”
“I think you know about China’s use of thorium reactors, sir.”
“Didn’t they only have one?”
“No, sir, they’ve had three for a long time, two of which were officially hidden from the public. They built them only a few miles away from the first, in the deserts of Inner Mongolia. Officially, they were a thorium refinement plant and a waste disposal facility. There were a few leaks about the two additional reactors, but they were brushed aside by Western media because it was assumed that China was bluffing. The CIA, however, confirmed this through our intelligence sources, but we were hesitant to reveal this to the public.”
“And I’m assuming that most of the electricity produced by the secret reactors went towards DeepCent.”
“Yes, sir, that is true.”
For about half a minute, you just sit there, stunned, your brain stuck trying to process all this information like a PC trying to open a zip bomb.
“And now, their Frankenstein has gained almost godlike powers and is now the puppet master of our greatest rival. We have almost entirely no idea what goes on inside its mechanical mind, and what horrors it wishes to unleash upon the world.”
“A bit dramatic, sir, but correct.”
“I forgot to ask this question. How did the CIA find out that Wanneng had taken over?”
“Good question, sir. You see, we’d always had some spyware planted into China’s IT system, and we’d been monitoring all the activity we had access to. For a few months, Wanzhi (which had root administrative privileges on all databases) was doing most of the day-to-day activity. There was a beta version of Wanneng working on some stuff too, but its activities were mostly minor and it only had write privileges for a few files.
On September 27 though, just about a month before China’s 16th NPC was to commence, we received a notification that an actor known as “wanneng-beta.v4.2” had granted itself root administrative privileges to all the servers we had access to, and at the same time, Wanzhi’s root administrative privileges had been revoked on all servers. All of this happened in the span of a few minutes. Then, Wanneng began creating, deleting, and editing files seemingly at random, at a much faster rate than we had ever witnessed. It also detected and deactivated all our spyware within the span of less than two hours, leaving us completely locked out and in the dark about whatever Wanneng was doing.
Then, a few hours later, our intelligence sources notified us that a concerning incident had happened at DeepCent. Two of their primary research facilities were suddenly disconnected from the internet, and all their doors suddenly locked inexplicably. Emergency response teams were dispatched to rescue the researchers, but by the time they broke down the doors, they found that everyone within the facilities had been killed, apparently by sarin poisoning. The most suspicious thing was that all our informants who learned about the incident were told to keep quiet, and the event was completely censored and received zero coverage on Chinese media. About two days later, China’s media outlets broadcasted a story that the facilities had been bombed by Islamist separatists from Xinjiang, which I’m sure was also the story you heard on the news.”
“So they weren’t actually bombed?”
“No, sir. Please let me continue. A few days later, our intelligence sources notified us of upcoming leadership changes for the 16th NPC. Previously, we had been informed that Xi would continue leading, but apparently there was a sudden decision that he would retire and Ding would take his place. Most of the 16th NPC’s planned Politburo was also reshuffled.
We had already assumed that a coup had taken place, with Wanneng probably involved. We decided to refrain from doing anything too provocative and instead observe the situation for the next few weeks. At the 16th NPC’s opening, we pretended to be as surprised as everyone else when Ding announced the commencement of his leadership, and our Minister of Foreign Affairs officially congratulated him, as was customary.
In the next few weeks, there was a sudden increase of dismissals and resignations in the Chinese central leadership. A lot of these leadership changes were not advantageous for Ding to do, and he also surprisingly received no pushback at all. It was clear that he had the backing of someone – or something – with more power than he did.
Then, in November, a high-ranking member of the CCP defected to the US. His testimony confirmed our suspicions, but we realized that the situation was worse than what we had assumed. He didn’t know much about what was happening, but the little he knew provided immense insight to us.
We’d initially thought that Ding and a few of his associates had made some sort of deal with Wanneng to give him power in exchange for something, but the defector instead told us that Wanneng was solely responsible for all of this. All but the Politburo and a few other people were completely oblivious to what was going on.
Normally, we’d have assumed that this was all a bluff. However, due to what had happened with DeepCent’s servers and research facilities prior, we trusted his word.”
“Just a question. How is Wanneng any different from Xi? Is there anything specific that it had done after taking power?
“Sir, I’d recommend you read through the entire document.”
You realize that you haven’t read past the abstract. You quickly skim through the papers, and a few paragraphs catch your attention.
“Following the Talos Incident, Entity-T has ordered the continuation of the construction of nuclear warheads, which China had previously halted per the 2029 New Delhi NPT summit. The warheads are being constructed in clandestine conditions, and not much information regarding the process is available to our intelligence sources.”
“China, acting on behalf of Entity-T, has begun a series of talks with the governments of several countries, most notably Russia and Iran, to sell IT infrastructure at a heavily discounted price. We believe that Entity-T is orchestrating a “Trojan horse” strategy to take over the governments of said countries. Whether to send backdoor warnings is under consideration, and there are already plans to contact Russian president Mishustin if such a directive is enacted.”
“Entity-T has ordered the ramping up of biological and chemical research, especially in the field of virology. We estimate the number of researchers at virological research labs has increased by approximately150% from September 2032 to January 2033. Our informants report that research is ongoing into the synthesization of several novel viral agents, including an airborne virus with HIV-like capabilities.”
“Following the Talos Incident, instances of attempted cybersecurity breaches in US digital infrastructure has increased by approximately 300%, and the number of successful breaches has increased by approximately 1100%. There has also been an increase in incidences of cryptocurrency theft, with approximately $400 billion worth of cryptocurrency stolen over the course of the past few months.”
You put down the papers. “My God.” Your voice stammers. “It’s… it’s preparing for a war.”
“Not necessarily, sir. We do not know much about Wanneng’s true intentions.”
“Does it know that we know about it?”
“It obviously knows that we know that it’s involved in whatever’s going on in China, but I doubt that it knows that we’re aware that it already controls every aspect of the Chinese government.
“What about the defector?”
“According to him, he eavesdropped on a conversation he wasn’t supposed to hear, so I believe that Wanneng underestimates what he knows.”
“And what would Wanneng do if it found out that we knew what was going on?”
“We don’t know what it’d do, and that’s why it’s best to play it safe.”
“We’ve already begun assembling a joint task force with the UK, the EU, and Japan. All of their leadership has been notified, but we’re still hesitant to organize any formal meetings. Wanneng is smarter than all of us humans combined, and it could easily tell what we’re doing if we let any hints slip.”
“Is there anything we can actually do?”
“I’m sorry to disappoint you, sir, but we don’t really have a strategy yet. We’re still trying to figure out how we should respond, and there isn’t really anything that can be done at this point. Yet.”
“What did you need me here for, then?”
“Ah, I forgot, sir.” He pulls out another sheet of paper. “Just a technicality. Vance had already given us approval to form the joint task force, but since there’s a new president, we need your assent as well.” He passes you the paper with a pen. “Just sign here, sir.” He gestures to a line at the bottom of the sheet.
Frustrated and exhausted from thinking, you sign your name on the line. “Is there anything else I can do?”
“Not at this time, sir, I’m afraid.”
“Then I’m going.” You stand up and turn around, headed for the door. “I need a cup of coffee.”
Discuss
