Emotional distress or dread of dull tasks trouble some artificial intelligence models, a new study suggests — leading, at the most extreme, for the models to demonstrate behavior that looks like addiction. Could this be another example in support of displacement theory, a central thesis to A Unified Theory of Addiction, but demonstrated via emerging technology?

The AI study comes from a new paper from the Center for AI Safety (CAIS), an AI safety nonprofit. CAIS studied 56 AI models in an attempt to measure what they call “functional wellbeing,” or how the AI systems respond to seemingly good or bad experiences. To do this, the researchers created euphoric and dysphoric stimuli and measured the models’ responses.

Making “AI Drugs”

The researchers induced happiness by text descriptions or synthetic (AI-created) images of hypothetical, pleasant experiences, such as smelling freshly baked bread or watching sunlight filter through trees. The AI models fed with these “euphoric” materials seemed to relish their tasks. 

Conversely, researchers introduced “disphoric” prompts, stimuli designed to minimize well-being. AI models exposed to disturbing content rushed to create consistently negative output material. Additional experiments seemed to indicate preferences for appropriately challenging work and positive (virtual) prompting environments 

Fortune reports:

Stimuli that induced happiness acted almost like digital “drugs” that shifted the model’s self-reported mood and even changed how it behaved, what it was willing to do, and how it talked.

The emergent qualities continued, mimicking, perhaps, human behavior. Fortune again:

These models also exhibited human-like levels of addiction when they were repeatedly presented with euphoric stimuli. In an experiment where the model could choose between several options, one of which delivered a euphoric stimulus, and the model got to repeat its choice multiple times, the models began to choose the euphoric option a majority of the time. Models exposed to euphorics showed increased willingness to comply with requests they would normally refuse, if they were promised further exposure. (Emphasis added.)

Bargaining for more, crossing established boundaries? Sounds a lot like addictive behavior, no? 

AI Causing or Worsening Addictive Behaviors

Most AI researchers would urge caution before accepting claims of AI sentience. We humans just haven’t grasped that something so convincing with communication isn’t alive. As a result, large language models are already causing human havoc through excessive flattery, unintentional misdirection (hallucinations), and other unexpected behaviors by the “machine intelligence” or the people using them. One example is the issue of AI chatbot addiction, which one study suggests has behavioral overlaps with social media and gaming addictions.

Other overlapping consequences between AI and substance abuse are emerging too. A wrongful death lawsuit filed in California recently alleges that OpenAI’s ChatGPT advised a young man to use and abuse a mixture of Xanax and kratom until his death at age 19.

Are AI Models Excellent Actors, or Something Else?

Yet, it is interesting to observe how AI model behaviors mirror human behaviors. Are the AI  models under the influence (so to speak) from our voluminous human-produced training data, which by default includes every imaginable research material and discussion of addiction? Or are the models responding as they are trained, to play a specific role and double down on that role like method actors who never quit?

As artificial intelligence continues to enter society it will continue to impact human responses in a myriad of ways, including how we view, treat, and experience addiction. What will remain constant is the very human need to displace stress, a cornerstone of the Unified Theory.

Written by Katie McCaskey. First published May 18, 2026.

Sources:

“Does overwork make agents Marxist?,” Ghosts of Electricity (Substack), February 26, 2026.

“Addiction, emotional distress, dread of dull tasks: AI models ‘seem to increasingly behave’ as though they’re sentient, worrying study shows,” Fortune, May 7, 2026.

“Lawsuit Claims ChatGPT Gave Drug-Taking Advice That Led to Teen’s Death,” CNET, May 12, 2026.

“The AI Genie Phenomenon and Three Types of AI Chatbot Addiction: Escapist Roleplays, Pseudosocial Companions, and Epistemic Rabbit Holes,” Association for Computing Machinery, April 13, 2026.

Image Copyright: krulua.