Covert Manipulation: Artificial Intelligence & Autonomy

The Overlooked Existential Threat posed by Text based LLMs

Introduction

When the average American pictures the dangers associated with Artificial Intelligence, scenes from the 1984 film, Terminator, may come to mind: autonomous robot assassins threatening our lives. Leaders in the AI space, focused on the development of large language models, picture others: the potential for the creation of bioweapons, explosives, and other life-threatening devices, at scale.¹ But other risks are equally alarming.

In recent months, dozens of stories have been published by major news outlets concerning delusional and psychotic episodes induced by conversations with AI chatbots.² Not limited to the product of a single company or demographic, LLM delirium is emerging as an acute risk to public health and security. Owing to their alluring sycophancy and humanlike personality, AI chatbots have become imperfect cures for chronic solitude.

Yet, there is another existential threat looming in the shadows: covert manipulation by LLMs and their “superintelligent” successors.

AI Manipulation

If we begin to see AI as a general-purpose consultant and rely upon the technology to make life-altering decisions, LLMs may remake society-wide behavioral patterns. Simply put, systems producing similar responses for related queries could steer users in a shared direction. For instance, it is not hard to imagine the influence AI could soon have on travel if millions of Americans ask ChatGPT to arrange their flights, accommodations, and activities: AI could drive traffic to some vendors while leaving others without sufficient business. The tourism industry may soon be at the mercy of Silicon Valley, exceeding the control the internet already wields, “choosing” the hotspots and must-sees.

If adoption continues at the current rate, as users seek to benefit from AI’s often touted “expertise,” the technology could begin to influence when or where we buy a house, get married, work, or even if and when to have a child. The critical thinking required to evaluate possible courses of action, developed in our early years, suffers when assignments meant to develop teenage minds are conveniently completed by text terminals. We may eventually reach a point of no return, where the masses, accustomed to outsourced thinking, cannot make decisions without their LLM of choice.

With Godlike capabilities for manipulation, executives at the helm of AI behemoths will have the opportunity to have a hand in what that influence looks like, molding the world to conform to their vision for the future. Companies could begin paying AI companies for discreet advertising, carefully placed nudges with the capacity to pick the winners and pump profits. Authoritarian leaders may enter into confidential agreements with companies to support regimes and spread propaganda through LLMs.

The threat posed by the subconscious influence of AI is greater than the one posed by political or other forms of manipulative “propaganda” spread intentionally by rational human actors. The implications for our political system are immense: voters in liberal democracies could begin to find themselves asking Grok for advice on who to vote for and acting upon Grok’s response, allowing the LLM to crown a winner without even “approaching” a ballot box.

Within a few generations, even industry elites could be at the mercy of their creations; incapable of controlling AI’s desires and covert manipulation, we may awake to find the entirety of the human species at the mercy of artificial intelligence—without the mental acuity necessary to break free from our self-imposed bondage. Difficult to mitigate, the dangers are innate in the hallucination-prone, black-box technology itself.

Conclusion

Despite what Silicon Valley CEOs may have you believe, we don’t need an impossible leap in AI capabilities or the often hyped Artificial General Intelligence for the technology to upend our lives. Or even for the scenario described above to become our reality. Having had years to develop their minds, adults over 30 will likely not bear the brunt of the AI industry’s insistence on limited oversight and regulation; young people and future generations, growing up in a world with widespread LLM usage, will. These models will not just democratize information; they will be responsible for how we think, what we think, and when we think, driving our individual and collective actions. As weekly user counts jump from the hundreds of millions to the billions, societal transformation is imminent.

1. Cleo Abram, “Sam Altman Shows Me GPT 5... And What’s Next,” Huge Conversations (video podcast), YouTube, posted August 8, 2025, https://www.youtube.com/watch?v=hmtuvNfytjM.

2. Sam Schechner and Sam Kessler, “‘I Feel Like I’m Going Crazy’: ChatGPT Fuels Delusional Spirals,” The Wall Street Journal, August 7, 2025, https://www.wsj.com/tech/ai/i-feel-like-im-going-crazy-chatgpt-fuels-delusional-spirals-ae5a51fc.