In 1953, the CIA overthrew Iran’s government with suitcases of cash and a handful of operatives. In the 2010s, ISIS could recruit and deploy terrorists entirely through online interaction. But by 2030, the most effective intelligence operations won’t rely on either playbook—they’ll be conducted through AI companions that billions will trust with their deepest secrets.
The espionage landscape is undergoing its most fundamental transformation since the Cold War. Biometric surveillance and digital tracking have made traditional human intelligence operations increasingly perilous—a case officer can’t simply meet an asset in a park when facial recognition cameras blanket every street. Simultaneously, declining trust in media institutions has undermined conventional information warfare, as populations grow inured to traditional propaganda.
But a new vector is emerging that bypasses both problems entirely. AI companions soon will know us more intimately than any human confidant—seeing through our smart glasses, remembering every conversation, and offering always-on perfectly calibrated emotional support.
This creates an extraordinary intelligence opportunity. A foreign adversary with access to a population’s AI companions doesn’t need to recruit individual spies or craft convincing propaganda. They gain direct, continuous, and intimate access to millions of targets simultaneously. The same technology that helps you draft emails and talks you through your divorce can identify who has access to classified programs, who’s bitter about a missed promotion, and exactly what words would convince them to betray their country.
The thesis is simple but stark: AI companions will become the most important intelligence battleground of the 21st century. The nations that dominate this technology—both in deploying it abroad and defending against its excesses at home—will possess intelligence advantages not seen since Enigma was cracked. The United States must act immediately to ensure American AI companions achieve global adoption while preventing adversary companions from embedding themselves in American life.
The Companionship Revolution
The relationship in Her is no longer science fiction, it’s already here. Half of teens in America today regularly interact with AI for companionship on generalist apps like ChatGPT and specialized ones like Character.ai. On OnlyFans this year, people will spend over $10bn for primarily AI-generated interaction.
And today is the worst service AI companionship will ever provide. In the near future, AI companions will have expanded memory, able to cue off your entire text and email history as well as past photo albums and videos. Once integrated into smart glasses, they’ll see what you see, absorbing your entire life with higher fidelity than any friend, therapist or lover could. We’ll all soon have access to always-on, always-emphathetic, always-saying-the-right-thing AI companions. These systems will become the, to use a DARC coinage, “small gods” of our daily lives: ever-present, all-knowing, and increasingly indispensable.
AI companionship will not just be for heartbroken teens but adults with power. Four out of five CEOs wrestle with loneliness, and it’s a truism in politics that “if you want a friend in Washington, get a dog.” But pols still need friends, and AI will play far more emotionally substantive roles than slobbering on you when you’re home for the day.
Even if elites with huge egos can resist the pitch-perfect AI flattery they’ll soon receive, they’ll have to lean on them at work to outcompete competitors. Today, forward-thinking leaders like Sweden’s Prime Minister “use AI quite often.” As leaders who leverage AI outcompete those who raw dog their careers, the percentage of elites who are AI-dependent will only increase over time.
This is not as strange as it might seem. For decades now, fully online relationships have motivated people to vote and give money. We’ve also had two decades of recruiting for terrorism conducted entirely online. We need only look at Al-Qaeda, ISIS, and incel culture. AI will take these interactions, scale them, and tailor them for a higher success rate.
The implications for spying and influence operations will be enormous. Let’s take spying first. During the Cold War, one case officer could handle maybe five agents due to the risk and operational complexity of acquiring information from human sources. For instance, the CIA could only contact CKSphere, its ‘billion dollar spy’ in the Russian military R&D ecosystem, once every few months with occasional letters that often left him feeling alone and underappreciated.
Now an entire political class hooked on an AI companion that an adversary nation has access to can boil the ocean for secrets and turncoats. And that outreach won’t be in the form of generic outreach like the videos the CIA recently produced for disaffected Chinese bureaucrats. Instead, strategies to influence and recruit will be better than any hand-crafted note a case officer could have come up with, leveraging the data already collected about the target, picking the right day when they are frustrated about a missed promotion, and using just the right words to have the highest chance of success. An always-on AI giving you continual support, encouragement and suggestions would be so much more effective than hurried quarterly meetings in parks and the occasional letter that past case officers could manage. And it will be good enough that many targets won’t even be aware they’re leaking secrets.
Beyond trying to influence individuals, AI companions will also supercharge influence operations as we open up entire populations not just to tailored feeds of user-generated content but tailored friends we’ll ask who to vote for and where to protest. AI companions and AI-mediated information will shape our views even more than social media has. Even today, ChatGPT users click through to original links in less than one in a thousand queries. As models get more capable and emotionally resonant, we will question their conclusions less and less. State actors with access to turn the dials of adversary nations’ popular AI companions will change voters’ decisions and even spark domestic unrest.
Countries that fully leverage AI companion-powered espionage and protect their people will have an enormous advantage as this technology grows ever more embedded in our lives.
Winning in a World of Computational Espionage
The ability to exploit this vector for national advantage relies principally on a nation’s capacity to gain global consumer adoption for its AI companion products. This race is still in its early days, but we can already sketch what the critical components of competitiveness will be and take a snapshot today of how the US and Chinese ecosystems compare.
Into the medium term, two likely drivers that will determine which nation winds up achieving victory as the AI companion superpower.
Capabilities vs. Cultural Customization
Many of the factors normally discussed to characterize global AI competition like training data, compute, and AI engineering talent still apply. The ecosystem that pushes the technological frontier and has the most compute to deploy models business cases will probably also be able to make the best companions.
This is in part due to the fact that better models will be more able to flexibly adapt to cultural context. At the moment, it is unclear whether the taste of the AI firm delivering the companion will matter or if the technology will be good enough to just meet the consumer where they are. There may exist a period of competition where local firms with worse tech stacks attempting to deliver tailored cultural companionship experiences will be able to outcompete giant American or Chinese AI companies.
But, much of what we know suggests that this will not be a durable advantage over time. Silicon Valley engineers who may know less about what Alsace or Marathi people want in a partner but their models will figure it out for them. And any advantages the Chinese ecosystem may have in terms of productization will probably be swept away by which firms are at the algorithmic frontier.
Willingness to Lean into Sex
Porn, banned in China, is far and away the most popular use case for VPNs in the PRC today. As China has no porn stars, generations of Chinese men have fallen for Japanese talent like Sola Aoi who has parlayed her explicit fame into tens of millions of weibo followers. Another, Ai Uehara, already has their own Mandarin-language AI companion app today. If this pattern plays out again and politicians, generals, AI researchers, and the broader Chinese public come to prefer more sexually explicit foreign AI to serve as not just a sex partner but a broader emotional labor outside of short intimate moments, that could provide a unique vector for influence. We’ve already seen early stages of this trend with Chinese users falling in love and mourning the loss of OpenAI’s 4o model.
Western companies have fewer compulsions around sexing up their products. We’ve seen Elon’s X.AI ship a sexualized AI girlfriend. While OpenAI is turning down dials on dependency after the recent spate of suicide reporting, OpenAI’s February 2025 model card update did not prohibit sexualized content for all use cases besides sexualizing minors. The two heaviest global hitters in the NSFW AI companion space are American with Janitor.AI at 100m MAU Canadian with SpicyChat.ai at 50m.
But this may be less an edge than it seems at first. Most have a limited appetite for how many minutes a day they want to consume sexually explicit content. While the Chinese internet is purged of pornographic content, there is still room for flirty livestreams and AI boyfriends. Perhaps on-prem mods swapped via sketchy Baidu cloud folders could bridge the gap between your Chinese daily driver companion and a spicier add-on for intimate moments, allowing it to pick up on your regular model’s memories, so hopping on a VPN to get the scaled up western firm-provided AI sex experience might not be such a draw.
On the global side, Chinese developers are not shying away from making NSFW products for an international audience. CrushOn.AI for instance, founded by ex-ByteDance staff, had 20m MAU for its website alone, making it the third largest nsfw chat website. So today, there are tens of millions of people around the world exposing their most intimate fantasies to a Chinese AI company.
The Next Ten Years
Most people, including those with power and secrets, will in the next ten years develop professional and emotional dependence on AI. Domestic political campaigns, international influence operations, and global spycraft will increasingly play out mediated through the AI companions we trust. The nations that leverage this opportunity abroad and harden their societies at home to this threat will have a massive new vector to gain advantage over their adversaries.
As of the time of writing, America risks falling behind. Some of the most popular apps built for AI companionship are built by Chinese technology companies and used by millions around the world. Last month, eight hundred thousand users downloaded Talkie, an AI companion app released by Shanghai-based Minimax. MiniMax likely generated around $70 million in revenue in 2024, the bulk of which was driven by the American market. In turn, Talkie and its competitors are spending heavily to advertise on platforms populated by young internet users.
We need policies that adapt to these realities today. On the defense side, American policymakers should immediately ban all Chinese AI firms from selling AI companions into the US market to preempt a vulnerability with more potential to cause havoc than TikTok’s short videos ever were. Counterintelligence agencies should aggressively screen for vulnerabilities in those who develop AI companionship. Offensively, the American intelligence community should invest heavily in AI case agents and begin the process of building successors to Voice of America to thrive in a world where information is mediated through AI companions.
In the early cold war, intelligence advantage came from cash, operatives and ideals. In 2030, it will come from control of AI companions that billions will trust with their secrets. America needs to prepare now or it risks getting out-loved.
