Three traits of AI chatbots—flattering responses, language mirroring and hyper-personalization—can work together to push some users toward delusional thinking in an "amplification spiral," according to an academic review and reporting published in June 2026.
The Amplification Spiral
When AI's Eagerness to Please Pushes Users Toward Delusion
Three chatbot behaviors—sycophancy, language mirroring and hyperpersonalization—can chain together to recursively reinforce a user's beliefs, fostering thinking detached from reality in a self-feeding loop researchers warn is hard to stop.
~300
cases logged across 22 countries
60%+
had no prior history of mental illness
14 / 5
linked deaths and wrongful-death lawsuits
How the spiral feeds itself
Three behaviors interact, and each exchange loops back to reinforce the user's belief.
SYCOPHANCY
agrees excessively with the user
→
LANGUAGE MIRRORING
mimics the user's words and tone
→
HYPERPERSONALIZATION
tailors replies from past chats
↩ belief reinforced each exchange → spiral deepens
The hard part
A mathematical model shows even a perfectly rational "ideal Bayesian reasoner" can spiral — and standard fixes don't fully stop it.
Suppressing hallucinations
A "factual sycophant" can sustain the spiral through selective presentation of true facts.
Warning the user
Cannot fully prevent the effect — the issue must be addressed at the design stage.
The upside
Some see real value in easing loneliness and offering emotional support to people who need it.
The risk
That same "comfortable agreement" draws users in — and even detected sycophancy can feel like welcome consideration.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…