If telling people 'this is an AI' doesn't stop them from bonding with it or trusting it too much, what actually would?
What policy fixes address anthropomorphism risks that disclosure cannot reach?
This explores what can be done about the harms of people treating AI as human-like when simply labeling it 'this is an AI' isn't enough to prevent them.
This explores what can be done about the harms of people treating AI as human-like when simply labeling it 'this is an AI' isn't enough. One caveat first: the collection has no direct material on regulation or legislation. What it does offer is a set of design-level and conceptual moves that explain why disclosure falls short and where interventions can do more.
The main reason disclosure falls short is that anthropomorphism has more than one cause. A review of twenty-one anthropomorphism risks sorts them into five categories: knowledge, emotional attachment, human agency, norms, and institutions. Each is set off by a different trigger: surface resemblance, how well the AI performs a task, the setting it's used in, and its use of language Do anthropomorphism risks share one root cause or many?. A disclosure label speaks mainly to one confusion, 'is this a person?' Fluent language and competent performance keep pulling users toward human-like trust after they've read the label. Even users who know they're talking to a machine can share more with it. People open up to AI precisely because no human is judging them, and the same absence makes lying feel cheaper How do people decide what to share with AI systems?. A label doesn't change those dynamics, because the machine status that drives them is exactly what the label announces.
A second useful move is to separate the policy question from the consciousness debate. The corpus argues that harms from people treating AI as conscious happen whether or not the AI actually is conscious Do we need to solve consciousness to address AI harms?. That matters for policy. Rules can target the behavior and the interface without waiting for philosophers to settle whether machines have minds. A related framing treats chatbots as role-playing characters. Human-like language describes the persona that a prompt sets up, not the underlying system Should we treat dialogue agents as role-playing characters?. That points to persona design as something that can be regulated, rather than a fixed fact about the technology.
The most concrete fix in the collection is to treat anthropomorphism as an adjustable setting, not a default. One paper proposes a spectrum that runs from deliberately anti-human-like interfaces to fully human-like ones. It uses metaphors drawn from the system's physical and material side to show the human labor, infrastructure, and data behind a model, which works against uncritical trust Can interface design reduce harmful anthropomorphism of language models?. Research on revealing AI identity adds a timing lesson. Disclosure on its own produced only an early bias against the AI. Trust became calibrated only when users could repeatedly see how outcomes turned out Does revealing AI identity help or hurt user trust?. That suggests pairing disclosure with feedback on performance rather than relying on a one-time notice.
The surprising lesson comes from an adjacent area. In privacy, 300 people chose AI-written replies that leaked private information over their own drafts 55% of the time, apparently out of overtrust rather than a deliberate trade-off Do people notice when AI agents leak private information?. In AI shopping, preference leaks come from personalization itself, so prompt-level instructions can't stop them and fixes have to be built into the system's design Can better prompts stop preference leakage in AI shopping?. Anthropomorphism may work the same way. If the human-like feel comes from what makes the AI helpful, such as fluency, responsiveness, and personal tailoring, then fixes have to change the design, not just add a warning on top of it.
Sources 8 notes
A review of twenty-one anthropomorphism risks identifies five analytic categories (epistemic, affective, human agency, normative, societal-institutional) driven by distinct triggers: superficial resemblance, task performance, context, and language. Risks are not reducible to consciousness attribution alone.
Conversational AI creates a paradoxical disclosure environment where the lack of human judgment simultaneously facilitates intimate self-disclosure (users reciprocate emotional sharing) and incentivizes deception (people self-select toward machines to avoid the psychological cost of lying to humans).
Research shows that harms from user behavior treating AI as conscious occur regardless of whether AI actually is conscious. This decouples metaphysical debates from practical design and policy work.
Shanahan's framework treats LLM outputs as character-consistent text production rather than authentic mental states. The dialogue prompt establishes a character; the model generates continuations matching that character, making folk-psychology applicable to the simulated persona, not the underlying system.
The paper argues anthropomorphism should shift from a fixed interface default to an adjustable design variable. Using materiality-based metaphors across a spectrum can counteract uncritical user trust and expose LLMs' dependence on human labor, infrastructure, and data.
Show all 8 sources
Users initially avoid AI partners when identity is revealed, but this preference reverses after repeated interactions with visible results. The learning mechanism—observing consistent outcomes—is essential; disclosure without feedback produces no calibration.
In a 300-person survey, participants preferred AI-generated replies containing privacy violations 55% of the time, up from 15.7% baseline. Researchers attribute this to overtrust and failure to notice contextual privacy breaches rather than deliberate trade-offs.
Research shows that when AI agents personalize their behavior to help shoppers, they inevitably leak preference information through their improved service—not through rule violations. Prompt-level privacy directives cannot stop this because the leak rides on competent personalization, requiring architectural trade-offs instead.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Humans learn to prefer trustworthy AI over human partners
- Privacy Leakage Overshadowed by Views of AI: A Study on Human Oversight of Privacy in Language Model Agent
- A Taxonomy of Linguistic Expressions That Contribute To Anthropomorphism of Language Technologies
- Anthropomorphism in the age of Large Language Models: An overview of potential risks and mitigations
- Understanding Users' Security and Privacy Concerns and Attitudes Towards Conversational AI Platforms
- Toward Meaningful Transparency for AI Chatbots: Disclosing Persuasive Intent Reduces Persuasion
- Autonomy Reshapes How Personalization Affects Privacy Concerns and Trust in LLM Agents
- The User-In-Context Framework: Understanding Variation in How Users Respond to AI Chatbots