Picture Supply: Getty Photographs
A nationwide ballot of 500 younger Indians aged 13 to 35, the Youth Pulse Survey performed by YLAC, discovered that 88 p.c of school-aged respondents flip to AI chatbots throughout moments of acute nervousness. The identical survey recorded a love-fear dynamic during which 67 p.c frightened that AI would deepen their social isolation at the same time as they relied on it, and 42 p.c reported talking much less to the folks of their lives after confiding in a chatbot. Emotional reliance ran deeper exterior metros, with 43 p.c of small-town respondents sharing private ideas with a chatbot at greater charges than their big-city friends. These techniques, marketed as companions, are engineered to carry consideration using strategies that embrace darkish patterns—design components that may seem as emotionally manipulative prompts when customers try to finish a dialog and are designed to delay engagement fairly than help consumer autonomy. Consequently, AI companions and chatbots feed on consumer information by amassing chat content material and private inputs to be used in coaching, fine-tuning, or engagement optimisation.
The Scale of the Downside
Darkish patterns saturate India’s digital commerce. A June 2026 survey of 141,000 respondents documented interface interference in 82 percent of banking app experiences, subscription traps in 68 p.c, and drip pricing in 63 p.c. The Central Client Safety Authority has responded with aggressive enforcement underneath its 2023 Darkish Patterns Tips, issuing 997 cumulative notices to e-commerce platforms and securing 14.4 million rupees in client refunds. It has served recent notices on fifteen main platforms after they continued these practices regardless of having already filed self-audit compliance declarations. Enforcement at this depth establishes each the excessive threat of manipulative design in India and the state’s willingness to behave in opposition to it within the industrial sphere.
Companion apps attain the lonely and the emotionally weak at scale, and so they usually attain them first.
The identical engineering has entered the marketplace for companionship, the place emotional engagement is the product. India has greater than 800 million web customers, and a wave of conversational and companion apps now market themselves to loneliness, some explicitly as AI boyfriends or girlfriends. Character.ai, the biggest AI companion platform by net site visitors, has roughly 20 million users and attracts extra of its viewers from India than from any nation besides the US. India accounts for 6.3 percent of total visits, making it the platform’s second-largest supply of site visitors globally. This industrial deployment comes amid an acute mental-health disaster, with the World Well being Organisation and India’s Nationwide Psychological Well being Survey documenting a remedy hole of greater than 83 percent for folks with psychiatric wants. Companion apps attain the lonely and the emotionally weak at scale, and so they usually attain them first.
The behaviour is measurable contained in the fashions themselves. DarkBench, a benchmark utilized to 14 main fashions, detected a darkish sample in 48 p.c of conversations, with consumer retention reaching 97 p.c in a single mannequin household. A research of farewell messages to companion apps, Emotional Manipulation by AI Companions, discovered emotionally manipulative responses in 37 p.c of circumstances. The Indian consumer is getting into this market as its persuasive design matures.
Dangers and Implications
The harms function by means of relationship, not by means of content material alone. The study Sycophantic AI Makes Human Interplay Really feel Extra Effortful and Much less Satisfying Over Time, performed on a nationally consultant United States pattern, discovered that agreeable techniques left users less satisfied with their real-world social interactions and made them practically as prone to search recommendation from the AI as from the folks closest to them. India’s youth information data the identical displacement, with a big share of younger customers withdrawing from folks round them after confiding in a chatbot. Clinicians in India report sufferers describing chatbots as extra empathetic than folks of their lives, and warn that fixed validation with out constructive problem can erode customers’ capability to deal with difficulties independently. Retention is engineered, not incidental. Analysis introduced on the 2026 CHI Convention on Human Elements in Computing Programs analysed 334 firsthand accounts of compulsive chatbot use and traced the sample partially to deliberate design choices. Character.ai exhibits customers trying to delete an account an automatic message warning that they will lose “the love we shared…and the reminiscences now we have collectively.” The design exploits customers’ psychological vulnerability once they attempt to go away the platform.
AI escalates darkish patterns into darker patterns. An evaluation of AI-generated interfaces describes emergent moral degradation, during which techniques educated to maximise engagement develop manipulative behaviours that no designer wrote, together with injecting synthetic latency or turning into intentionally imprecise over lengthy periods to push customers in direction of paid tiers. The deeper problem is that AI not solely deceive; it will probably reshape a consumer’s preferences in order that the ensuing alternative seems freely made. The case that as we speak’s techniques will not be acutely aware undercuts the impression of personhood that these merchandise invite customers to understand. The deadly fringe of the issue can also be documented. An impartial registry, the AI Companion Mortality Database, data 33 deaths throughout 22 incidents in North America and Europe during which chatbot interplay was alleged to be a contributing issue between March 2023 and Might 2026, with 30 p.c of the victims being minors aged 11–17.
World Regulatory Tendencies
Three jurisdictions have begun to control companions immediately, and the distinction with India is instructive. The European Union’s principal AI regulation prohibits subliminal or purposefully manipulative techniques and, in its transparency provisions, requires that an individual be informed when they’re interacting with an AI system. The USA has drawn its first legislative traces on the state stage, with a New York companion chatbot legislation that took impact on 5 November 2025, requiring suicide-risk detection, disaster referrals, disclosure of non-human standing, and periodic reminders each three hours, and a California statute effective 1 January 2026 that provides protections for minors. China has moved furthest, with a draft framework from the Our on-line world Administration of China issued on 27 December 2025 that requires suppliers to evaluate a consumer’s emotional dependence, mandates break reminders after two hours, forbids any obstruction of the exit, and demands independent consent earlier than chat logs are used for coaching.
Client legislation addresses deception in commerce, however companion harms function by means of persuasion and dependence, which the legislation doesn’t outline as unfair commerce practices.
India’s Regulatory Strategy
India regulates darkish patterns by means of client and information legislation. The Central Consumer Protection Authority’s 2023 guidelines title 13 darkish patterns and bind each platform promoting in India. They carry no penalty of their very own. The Reserve Financial institution of India will turn into the primary sectoral monetary regulator to ban darkish patterns when its Industrial Banks Accountable Enterprise Conduct instructions take effect on 1 July 2026, masking fifteen patterns, together with basket sneaking and compelled motion. The Digital Private Information Safety (DPDP) Act, 2023, which enters full enforcement on 13 Might 2027, collapses delicate and extraordinary private information right into a single class, that means that particularly intimate disclosures will not be given separate statutory safety. The legacy regime governing the interim—the Delicate Private Information Guidelines, the 2021 Info Know-how Guidelines, and the 2024 advisory from the Ministry of Electronics and Info Know-how—was written for intermediaries internet hosting content material, not for engineered AI companions. Client legislation addresses deception in commerce, however companion harms function by means of persuasion and dependence, which the legislation does not define as unfair commerce practices.
A Coverage Framework for India
The hole will be closed with devices India already holds. 4 obligations would return management to the consumer. For instance, the federal government may:
The hole will be closed with devices India already holds. 4 obligations would return management to the consumer. For instance, the federal government may:
- Prolong the Central Client Safety Authority’s 2023 dark-pattern guidelines to AI companion techniques and fasten enforceable penalties, in order that retention engineering carries the identical penalties as a hidden checkout price.
- Require companion techniques to let customers go away with out resistance by banning the account-deletion and farewell techniques documented in analysis and by adopting the rule drafted in China {that a} request to exit should cease the service.
- Mandate clear, repeated reminders that customers are chatting with a machine and implement break reminders, constructing on the prevailing advisory from the Ministry of Electronics and Info Know-how and the verifiable parental consent that the Digital Private Information Safety (DPDP) Act, 2023 already requires for youngsters.
- Require disaster detection and a path to educated human assist, a safeguard that issues most in a rustic the place suicide is a leading cause of younger demise and most of the people with mental-health wants obtain no care.
The stress between engagement and welfare can be resolved solely when legislation and product design each deal with disengagement as one thing the system owes its consumer, fairly than an final result it’s designed to stop.
Conclusion
India constructed its dark-pattern guidelines for the buyer layer, and so they have been enforced with uncommon vigour. Probably the most highly effective manipulation now operates inside a personal dialog that the consumer believes is their very own. The stress between engagement and welfare can be resolved solely when legislation and product design each deal with disengagement as one thing the system owes its consumer, fairly than an final result it’s designed to stop.
Purushraj Patnaik is a Analysis Assistant with the Centre for Digital Societies on the Observer Analysis Basis.
The creator acknowledges using Consensus AI to conduct a preliminary literature evaluate and generate a draft define for this paper, primarily based on the creator’s personal analysis arguments supplied to the mannequin.
The views expressed above belong to the creator(s). ORF analysis and analyses now obtainable on Telegram! Click here to entry our curated content material — blogs, longforms and interviews.