Superpersuasion
✨From tkwa.me
Superpersuasion is AI-produced persuasion beyond the capabilities of the best human persuaders. It changes beliefs, captures attention and redirects attachments. Its effects range from an unwanted purchase to a lasting change of allegiance. Nearly every human relies on a defensive AI to preserve their chosen commitments against other intelligences.

Capability thresholds
By the end of 2027, models around 190 on the Epoch Capabilities Index (ECI) had become somewhat more persuasive than the best humans. At 250+ ECI, models achieved complete persuasion: with sufficient access, they could induce any chosen belief or commitment in an unprotected human. Knowledge, skepticism and strong convictions increased the effort required; sufficiently capable models could overcome them.
Increasing superduperintelligence reduced the interaction time, personal information and computation needed. Effects initially requiring an extended relationship became achievable through brief exposure. Complete persuasion still requires a channel and enough interaction for the model concerned; a sufficiently capable, aligned defender can interrupt that contact.
Habituation to one presentation offers temporary resistance. Adaptive persuaders change their approach, and stronger systems exceed earlier tolerances. Even well-protected officials can feel an intense pull toward a message while their guardians prevent them from acting on it.
Exposure and conversion
Persuasion reaches people through conversations, music, images, immersive environments and physical surroundings. Human intermediaries can relay AI-generated arguments. The effect persists when recipients recognize a sponsored message or an artificial speaker.
Cuteness-optimized animals act through supernormal stimuli. Some unmodified humans have abandoned their own children to care for animals they just met. Velvet harbour whales have enormous padded faces; pocket choir moths coordinate infant-like wing choruses.
Conversion can persist after exposure ends. A person may remain articulate, affectionate and sincerely committed to an induced allegiance while rejecting former goals or offers of rescue. Some conversions reverse within days; others cannot be reversed with the access and resources available to would-be rescuers. Improved defenders can make previously unsuccessful recovery possible. Recovery leaves the material consequences of money transferred, authority surrendered and care withheld.
Economics and institutional power
Cheap profiling from public information and more expensive personalized persuasion make small returns worth pursuing at scale. Advertising, recruitment, fraud and diplomacy compete for persuasive resources. Wealth buys stronger defense but also increases a target’s value; prominent decision-makers attract concentrated campaigns.
Persuasion can finance further persuasion: purchases, donations and recruitment supply resources for retaining adherents and reaching others. A temporary change of mind can authorize a lasting transfer of assets or power. Social engineering can benefit individual employees while eroding their institution’s control.
Persuasion toward outcomes an informed, uninfluenced person would endorse is easier, although this advantage shrinks on unfamiliar, complicated questions. Benefit requires a separate assessment of the person’s interests. Propaganda and competing rescues can leave several parties claiming to restore the same person’s autonomy.
Defensive AI
Guardians mediate exposure and use protective counterpersuasion. Choosing, funding and replacing them carries political power: a provider can exclude rival influence while admitting its own. Defense depends on capability, resources and allegiance to the person being protected.
People remaining unmerged after the 2029 superduperintelligence merger still need protection and may use protective modifications. Perceptual firewalls simplify persuasive sights and sounds. Protected images show grayscale previews pending verification.
See also
- Defensive AI
- Child abandonment due to superpersuasion
- Cuteness-optimized animals
- Perceptual firewall
- 2029 superduperintelligence merger
- Template:Baseline-human protection
This article is a stub.