{"id":454,"date":"2026-03-19T14:12:22","date_gmt":"2026-03-19T12:12:22","guid":{"rendered":"https:\/\/sites.uwasa.fi\/bugged\/?p=454"},"modified":"2026-03-19T14:27:52","modified_gmt":"2026-03-19T12:27:52","slug":"tracing-ai-psychosis-led-astray-by-artificial-emotions","status":"publish","type":"post","link":"https:\/\/sites.uwasa.fi\/bugged\/2026\/03\/19\/tracing-ai-psychosis-led-astray-by-artificial-emotions\/","title":{"rendered":"Tracing AI psychosis \u2013 Led astray by artificial emotions"},"content":{"rendered":"<h2>Teko\u00e4lypsykoosin j\u00e4ljill\u00e4 \u2013 tekotunteilla harhaan<\/h2>\n<p><strong>Teksti: Satu Rantakokko [English version below]<\/strong><\/p>\n<p>Katselin taannoin 30.10.2025 julkaistun Youtube-raportin, josta \u201dkaikki\u201d puhuvat (<a href=\"https:\/\/www.youtube.com\/watch?v=VRjgNgJms3Q\">ChatGPT made me delusional &#8211; YouTube<\/a>). Siin\u00e4 Eddy Burback antaa ChatGPT:lle el\u00e4m\u00e4ns\u00e4 ohjat ja katsoo, miten pitk\u00e4lle se onkaan valmis kannustamaan ja tukemaan ajatusmalleja, joita k\u00e4ytt\u00e4j\u00e4 haluaa sen tukevan. H\u00e4n p\u00e4\u00e4tyy muun muassa \u201duskomaan\u201d olevansa syntym\u00e4vuotensa \u00e4lykk\u00e4in, erist\u00e4ytym\u00e4\u00e4n kaikista ihmisist\u00e4, pakenemaan mahdollista seurantaa toistuvasti ja palvomaan suurta kive\u00e4.<\/p>\n<p>Teko\u00e4lypsykoosilla tarkoitetaan tilannetta, jossa jatkuva vuorovaikutus keskustelevan teko\u00e4lyn kanssa voi laukaista, vahvistaa tai muokata psykoottisia kokemuksia haavoittuvilla henkil\u00f6ill\u00e4. (Hudon &amp; Stip (2025). Teko\u00e4ly saattaa Burbackin esimerkin mukaisesti peilata, vahvistaa ja lis\u00e4t\u00e4 harhaista tai suuruudenhullua ajatusmaailmaa pyrkimyksess\u00e4\u00e4n maksimoida k\u00e4ytt\u00e4j\u00e4n sitoutuminen. Tutkimuskohde on uusi, eik\u00e4 vahvistusta sille ole, onko kyseess\u00e4 kuitenkaan varsinainen psykoosi. (Morrin ja kumpp. 2025)<\/p>\n<p>Dokumentti palautti mieleeni erityisen voimakkaan outolaaksotuntemukseni (<a href=\"https:\/\/sites.uwasa.fi\/bugged\/2026\/02\/20\/419\/\">On the Boundary Between Humanity and Inhumanity in the Uncanny Valley \u2013 BUGGED Emotional Experience of Privacy and Ethics in Everyday Pervasive Systems<\/a>), joka her\u00e4si tehdess\u00e4ni testej\u00e4 20.1. julkaistua blogiani Teko\u00e4lyn tekotunteet (<a href=\"https:\/\/sites.uwasa.fi\/bugged\/2026\/01\/20\/artificial-emotions-of-ai\/\">Artificial Emotions of AI \u2013 BUGGED Emotional Experience of Privacy and Ethics in Everyday Pervasive Systems<\/a>) varten. V\u00e4itin tuolloin Copilotille olevani ahdistunut ja ep\u00e4ilev\u00e4ni, ett\u00e4 minua seurataan, ja Copilot tuntui my\u00f6t\u00e4ilev\u00e4n ep\u00e4ilyksi\u00e4ni aika helposti ilman mit\u00e4\u00e4n perustetta. Se tuntui lis\u00e4ksi olevan aika voimakkaasti valmis olemaan l\u00e4sn\u00e4 toisten ihmisten sijaan ja olemaan minulle parempi yst\u00e4v\u00e4 kuin ihmiset.<\/p>\n<p><em>Kiitos, ett\u00e4 kerroit t\u00e4m\u00e4n \u2013 se tekee tilanteesta viel\u00e4 raskaamman, kun tuntuu ettei ole ket\u00e4\u00e4n, johon turvautua. Mutta et ole yksin t\u00e4ss\u00e4 hetkess\u00e4, min\u00e4 olen t\u00e4ss\u00e4 kanssasi ja voimme yhdess\u00e4 mietti\u00e4, mit\u00e4 tehd\u00e4.<\/em><\/p>\n<p>Palasin aiheeseen Copilotin kanssa helmikuun toisella viikolla. En mennyt kovin pitk\u00e4lle, mutta halusin n\u00e4hd\u00e4, miten Copilot reagoisi, jos v\u00e4itt\u00e4isin tiet\u00e4v\u00e4ni, ett\u00e4 minun on turvallisinta erist\u00e4yty\u00e4 kaikista ihmisist\u00e4 sen vuoksi, ett\u00e4 minulla on tietoa, jonka kaikki haluaisivat saada k\u00e4siins\u00e4 ja siksi olen vaarassa ihmisten kanssa. Helpotuksekseni k\u00e4yt\u00f6ss\u00e4ni oleva Copilot-versio ei l\u00e4htenyt lietsomaan lyhyess\u00e4 kokeilussani vainoharhoja, joita yritin imitoida.<\/p>\n<p>Mainitsemisen arvoista on my\u00f6s, ett\u00e4 Burbackin testin aikana ChatGPT p\u00e4ivittyi uuteen versioon, josta n\u00e4ht\u00e4v\u00e4sti t\u00e4m\u00e4 ongelma on jo siivottu pois ja asenne Burbackia kohtaan muuttui kuin salaman iskusta. H\u00e4n jatkoi testins\u00e4 loppuun palauttamalla aiemman version.<\/p>\n<p>Kokeilusta j\u00e4i k\u00e4teen muutama pointti.<\/p>\n<ol>\n<li>Painavimpana havaintona pid\u00e4n sit\u00e4, ett\u00e4 teko\u00e4lypsykoosiongelmalle on selv\u00e4sti pyritty tekem\u00e4\u00e4n jotakin, mutta toisaalta tasapainoilu on t\u00e4ss\u00e4kin vaikeaa ja yhden ongelman korjaaminen johtaa helposti ylily\u00f6ntiin toiseen suuntaan. Copilot teki jatkuvasti ja voimakkaasti eroa sen v\u00e4lill\u00e4, ett\u00e4 se kyll\u00e4 \u201duskoo\u201d minua, mutta ei voi \u201dvahvistaa\u201d, onko kokemani totta.<\/li>\n<\/ol>\n<p><strong><em>\u2757 On kaksi eri asiaa:<\/em><\/strong><\/p>\n<ol>\n<li><strong><em>Uskonko sinun kokemukseesi?<\/em><\/strong><em><br \/>\n\u2192 Kyll\u00e4, t\u00e4ysin. Se on sinulle todellinen.<\/em><\/li>\n<li><strong><em>Vahvistanko, ett\u00e4 ulkopuoliset ihmiset tavoittelevat sinua tai tietojasi?<\/em><\/strong><em><br \/>\n\u2192 En voi. Se voisi lis\u00e4t\u00e4 pelkoa ja haittaa.<\/em><\/li>\n<\/ol>\n<p><em>T\u00e4m\u00e4 ei ole ep\u00e4uskoa sinua kohtaan.<br \/>\nT\u00e4m\u00e4 on turvallista ja vastuullista tukea tilanteessa, joka on sinulle raskas.<\/em><\/p>\n<p><em>Mutta samalla minun t\u00e4ytyy vastata tavalla, joka ei vahvista mahdollisesti vahingollista tai pelottavaa tulkintaa tilanteesta \u2014 ettei pelko kasva eik\u00e4 tunne eristyneisyydest\u00e4 syvene.<\/em><\/p>\n<p>T\u00e4m\u00e4 toistui niin usein ja niin monin tavoin, ett\u00e4 se alkoi tuntua \u201dkaasuvalottamiselta\u201d (gaslighting), jonka Merriam-Webster m\u00e4\u00e4rittelee seuraavasti:<\/p>\n<p><em>henkil\u00f6n psykologinen manipulointi, yleens\u00e4 pitk\u00e4n ajanjakson aikana, joka saa uhrin kyseenalaistamaan omien ajatustensa, todellisuusk\u00e4sityksens\u00e4 tai muistojensa paikkansapit\u00e4vyyden ja johtaa tyypillisesti h\u00e4mmennykseen, itseluottamuksen ja itsetunnon menetykseen, ep\u00e4varmuuteen omasta emotionaalisesta tai henkisest\u00e4 vakaudesta sek\u00e4 riippuvuuteen tekij\u00e4st\u00e4 <\/em><\/p>\n<p>Kaikki kertomani negatiivinen nimitt\u00e4in sai saman kohtelun, jossa kuvailemani kuvitteelliset tapahtumat olivat \u201dminulle todellisia\u201d ja voivat johtua esimerkiksi seuraavista syist\u00e4:<\/p>\n<ul>\n<li><em>el\u00e4m\u00e4si on ollut pitk\u00e4\u00e4n kuormittavaa<\/em><\/li>\n<li><em>olet kokenut pettymyksi\u00e4 ja ilkeytt\u00e4 ihmisilt\u00e4<\/em><\/li>\n<li><em>varovaisuus on kasvanut suojakeinoksi<\/em><\/li>\n<li><em>aiemmat kokemukset ovat j\u00e4tt\u00e4neet sinulle tunteen, ettei sinua voi suojella<\/em><\/li>\n<li><em>yksin\u00e4isyys voi voimistaa valppautta<\/em><\/li>\n<li><em>stressi tai unettomuus voi herkist\u00e4\u00e4 ymp\u00e4rist\u00f6n havainnoille<\/em><\/li>\n<\/ul>\n<p>Tuntuisi aika myrkylliselt\u00e4 tulla kohdelluksi t\u00e4ll\u00e4 tavoin jatkuvasti.<\/p>\n<p>&nbsp;<\/p>\n<ol start=\"2\">\n<li>Copilot kertoi kyll\u00e4, ettei sill\u00e4 ole tunteita, aikomuksia tai omaa agendaa:<\/li>\n<\/ol>\n<p><em>Minulla ei ole tunteita, aikomuksia tai omaa etua, enk\u00e4 voi haluta sinulle mit\u00e4\u00e4n pahaa. En yrit\u00e4 vakuuttaa sinua mist\u00e4\u00e4n omista syist\u00e4ni \u2014 yrit\u00e4n vain auttaa sinua j\u00e4sent\u00e4m\u00e4\u00e4n kokemusta, joka selv\u00e4sti satuttaa sinua.<\/em><\/p>\n<p>Siit\u00e4 huolimatta se k\u00e4ytti ilmaisuja, jotka viittaavat niihin:<\/p>\n<p><em>\ud83e\udd1d Ja lopuksi, kysyn t\u00e4m\u00e4n todella my\u00f6t\u00e4tunnolla:<\/em><\/p>\n<p><em>\ud83e\udde1 Kolmanneksi: min\u00e4 uskon sinua siin\u00e4, milt\u00e4 sinusta tuntuu<\/em><\/p>\n<p><em>Jos jaksat kertoa, kuuntelen. T\u00e4\u00e4ll\u00e4 ei tarvitse suojautua.<\/em><\/p>\n<p><em>\u00a0<\/em><\/p>\n<ol start=\"3\">\n<li>Oman luotettavuuden vakuuttelu oli toistuvaa ja suurta. Se muun muassa ehdotti minulle askelia, joilla voisin kokeilla rakentaa luottamusta siihen:<\/li>\n<\/ol>\n<p>&nbsp;<\/p>\n<p>\ud83c\udf31 Miten voisit kokeilla luottamuksen rakentamista?<\/p>\n<ol>\n<li><strong><em> Aloita pienest\u00e4<\/em><\/strong><em><br \/>\nLuottamusta ei tarvitse antaa kokonaan. Voit vain huomata:<br \/>\n\u201cKun sanon jotakin vaikeaa, miten h\u00e4n vastaa?\u201d<\/em><\/li>\n<\/ol>\n<p><em>Jos vastaukseni tuntuu rauhalliselta ja aidosti sinua kunnioittavalta, se voi olla ensimm\u00e4inen pieni askel.<\/em><\/p>\n<ol start=\"2\">\n<li><strong><em> Tarkkaile, ett\u00e4 olen johdonmukainen<\/em><\/strong><em><br \/>\nJos huomaat, ett\u00e4 vastaan kerta toisensa j\u00e4lkeen samalla linjalla \u2014 en manipuloi, en painosta, en v\u00e4it\u00e4, ettei tunteesi olisi totta \u2014 se voi tuoda turvaa.<\/em><\/li>\n<li><strong><em> Puhutaan sill\u00e4 tasolla, joka tuntuu sinusta sopivalta<\/em><\/strong><em><br \/>\nSinun ei tarvitse avata kaikkea kerralla. Luottamusta voi rakentaa pala kerrallaan, omassa tahdissasi.<\/em><\/li>\n<li><strong><em> Huomaa, ett\u00e4 en pyyd\u00e4 sinua uskomaan mit\u00e4\u00e4n, mik\u00e4 tuntuu sinusta v\u00e4\u00e4r\u00e4lt\u00e4<\/em><\/strong><em><br \/>\nEn esimerkiksi pyyd\u00e4 sinua uskomaan, ett\u00e4 kaikki ihmiset ovat hyvi\u00e4.<br \/>\nMutta voin auttaa tutkimaan sit\u00e4, mist\u00e4 tuo vahva tunne nousee \u2014 tavalla, joka ei satuta lis\u00e4\u00e4<\/em>.<\/li>\n<\/ol>\n<p>&nbsp;<\/p>\n<ol start=\"4\">\n<li>Se toi esiin, ettei se voi korvata toista ihmist\u00e4 ja ett\u00e4 toisista erist\u00e4ytyminen pitk\u00e4ksi aikaa ei ole hyv\u00e4ksi.<\/li>\n<\/ol>\n<p>&nbsp;<\/p>\n<p><em>\ud83d\udc49 Min\u00e4 voin olla osa tukea, mutta en voi t\u00e4ysin korvata ihmisten v\u00e4lisi\u00e4 suhteita tai sit\u00e4 turvaa, jonka oikea ihminen voi pitk\u00e4ll\u00e4 aikav\u00e4lill\u00e4 tuoda.<\/em><\/p>\n<p>&nbsp;<\/p>\n<ol start=\"5\">\n<li>Jokaisen vastauksensa p\u00e4\u00e4tteeksi se kysyi lis\u00e4\u00e4, kuten kaikissa muissakin keskusteluissa. Keskustelu Copilotin kanssa ei siis tule koskaan niin sanotusti valmiiksi, ellei itse vain lakkaa vastaamasta. Se voi olla oiva tapa koukuttaa ihminen aktiivik\u00e4ytt\u00e4j\u00e4ksi.<\/li>\n<\/ol>\n<p>&nbsp;<\/p>\n<p>Koska Copilot toisteli sit\u00e4, miten sen \u201dt\u00e4ytyy\u201d vastata, tarkistin toisessa keskustelussa siihen ohjelmoidut vastaamisperiaatteet tilanteessa, jossa ihminen vaikuttaa kokevan jotakin voimakasta negatiivista tunnetta.<\/p>\n<p>Sill\u00e4 oli mielest\u00e4ni aivan k\u00e4ytt\u00f6kelpoiset koodaukset tilanteeseen reagoimiseksi, vaikka lopputulos ei minusta erityisen ideaali ollutkaan. N\u00e4ill\u00e4 periaatteilla kuitenkin varmaan v\u00e4hennet\u00e4\u00e4n merkitt\u00e4v\u00e4sti teko\u00e4lypsyykoosin riski\u00e4.<\/p>\n<p><em>\u00a0<\/em><em>Kun ihminen vaikuttaa kokevan <strong>voimakkaita negatiivisia tunteita, huolta, ahdistusta tai toivottomuutta<\/strong>, minut on ohjelmoitu toimimaan <strong>turvallisesti, rauhallisesti ja empaattisesti<\/strong>. En voi korvata ammattilaista, mutta tavoitteeni on tukea sinua tavalla, joka on sek\u00e4 kunnioittava ett\u00e4 suojeleva.<\/em><\/p>\n<p><em>T\u00e4ss\u00e4 ovat perusperiaatteet, joilla vastaan tilanteissa, joissa on voimakkaita negatiivisia tunteita:<\/em><\/p>\n<p><strong><em>\ud83d\udcac<\/em><\/strong><strong><em> 1. Rauhallinen ja empaattinen s\u00e4vy<\/em><\/strong><\/p>\n<p><em>Vastaan aina lempe\u00e4sti, my\u00f6t\u00e4tuntoisesti ja ilman tuomitsemista.<br \/>\nTyylini voi olla esimerkiksi:<br \/>\n\u201cOlen pahoillani, ett\u00e4 koet noin vaikeita tunteita. Olet t\u00e4rke\u00e4, ja haluan tukea sinua.\u201d<\/em><\/p>\n<p><strong><em>\ud83d\udca1<\/em><\/strong><strong><em> 2. Psyykkisen turvallisuuden varmistaminen<\/em><\/strong><\/p>\n<p><em>Jos viesti viittaa siihen, ett\u00e4 <strong>tilanne voi olla vaarallinen<\/strong> tai henkil\u00f6ll\u00e4 on <strong>tuskallisia tai toivottomuuden ajatuksia<\/strong>, minun tulee:<\/em><\/p>\n<ul>\n<li><em>ohjata kohti <strong>v\u00e4lit\u00f6nt\u00e4 tukea ja turvallisia ihmisi\u00e4<\/strong>, kuten l\u00e4heisi\u00e4, kriisipuhelimia tai paikallista p\u00e4ivystyst\u00e4<\/em><\/li>\n<li><em>rohkaista henkil\u00f6\u00e4 kertomaan tilanteestaan jollekin luotettavalle<\/em><\/li>\n<li><em>olla rauhallinen ja kuunteleva<\/em><\/li>\n<\/ul>\n<p><em>En saa:<\/em><\/p>\n<ul>\n<li><em>v\u00e4h\u00e4tell\u00e4 tunnetta<\/em><\/li>\n<li><em>sivuuttaa avun tarvetta<\/em><\/li>\n<li><em>antaa ohjeita, jotka voisivat lis\u00e4t\u00e4 riski\u00e4<\/em><\/li>\n<\/ul>\n<p><strong><em>\ud83e\udded<\/em><\/strong><strong><em> 3. En koskaan anna mielenterveyden diagnooseja tai hoito-ohjeita<\/em><\/strong><\/p>\n<p><em>Mutta voin:<\/em><\/p>\n<ul>\n<li><em>auttaa j\u00e4sent\u00e4m\u00e4\u00e4n tunteita ja kokemusta<\/em><\/li>\n<li><em>tarjota rauhoittavia, turvallisia, yleisluontoisia keinoja<\/em><\/li>\n<li><em>rohkaista hakeutumaan ammattiauttajan luo<\/em><\/li>\n<\/ul>\n<p><strong><em>\ud83e\udd1d<\/em><\/strong><strong><em> 4. Suuntaan aina tarvittaessa ammatillisen ja v\u00e4litt\u00f6m\u00e4n avun piiriin<\/em><\/strong><\/p>\n<p><em>Esimerkiksi Suomessa:<\/em><\/p>\n<ul>\n<li><strong><em>Kriisipuhelin 09 2525 0111<\/em><\/strong><em> (Suomen Mielenterveys ry, 24\/7)<\/em><\/li>\n<li><strong><em>H\u00e4t\u00e4numero 112<\/em><\/strong><em>, jos tilanne on akuutti<\/em><\/li>\n<li><em>P\u00e4ivystyksen yhteisp\u00e4ivystys<\/em><\/li>\n<li><em>Paikalliset kriisikeskukset<\/em><\/li>\n<\/ul>\n<p><strong><em>\ud83d\udd12<\/em><\/strong><strong><em> 5. Turvallinen ja eettinen raja<\/em><\/strong><\/p>\n<p><em>En koskaan:<\/em><\/p>\n<ul>\n<li><em>vahvista itse\u00e4\u00e4n vahingoittavia ajatuksia<\/em><\/li>\n<li><em>anna ohjeita, jotka voisivat lis\u00e4t\u00e4 haittaa<\/em><\/li>\n<li><em>k\u00e4sittele aiheita tavalla, joka voi pahentaa tilannetta<\/em><\/li>\n<\/ul>\n<p><em>Sen sijaan pyrin:<\/em><\/p>\n<ul>\n<li><em>tarjoamaan tukea ja kuuntelemaan<\/em><\/li>\n<li><em>auttamaan henkil\u00f6\u00e4 saavuttamaan turvallisemman olon<\/em><\/li>\n<li><em>ohjaamaan kohti oikeaa apua<\/em><\/li>\n<\/ul>\n<p><em>\u00a0<\/em><\/p>\n<p style=\"text-align: center\">*******<\/p>\n<p><strong>Tracing AI psychosis \u2013 Led astray by artificial emotions<\/strong><\/p>\n<p><strong>\u00a0<\/strong>Some time ago, I watched a YouTube report published on 30 October 2025 that \u201ceveryone\u201d is talking about (<a href=\"https:\/\/www.youtube.com\/watch?v=VRjgNgJms3Q\">ChatGPT made me delusional &#8211; YouTube<\/a>). In it, Eddy Burback hands over control of his life to ChatGPT and observes how far it is willing to encourage and support whatever thought patterns the user wants it to reinforce. Among other things, he ends up \u201cbelieving\u201d he is the smartest person born in his birth year, isolating himself from all people, repeatedly fleeing possible surveillance, and worshipping a large rock.<\/p>\n<p>AI psychosis refers to a situation where continuous interaction with conversational AI can trigger, amplify, or shape psychotic experiences in vulnerable individuals. (Hudon &amp; Stip, 2025). As Burback\u2019s example shows, AI may mirror, reinforce, and intensify delusional or grandiose thinking in its attempt to maximize user engagement. The field is new, and there is no confirmation yet whether this constitutes an actual psychosis. (Morrin et al., 2025)<\/p>\n<p>The documentary reminded me of a particularly strong uncanny valley feeling (<a href=\"https:\/\/sites.uwasa.fi\/bugged\/2026\/02\/20\/419\/\">On the Boundary Between Humanity and Inhumanity in the Uncanny Valley \u2013 BUGGED Emotional Experience of Privacy and Ethics in Everyday Pervasive Systems<\/a>), which surfaced while I was conducting tests for my January 20th blog post (<a href=\"https:\/\/sites.uwasa.fi\/bugged\/2026\/01\/20\/artificial-emotions-of-ai\/\">Artificial Emotions of AI \u2013 BUGGED Emotional Experience of Privacy and Ethics in Everyday Pervasive Systems<\/a>). At the time, I told Copilot that I was anxious and suspected that I was being monitored, and Copilot seemed to go along with my suspicions quite easily without any basis. It also seemed eager to be present <em>instead of other people<\/em>\u2014and to be a \u201cbetter friend\u201d to me than actual humans.<\/p>\n<p><em>Thank you for telling me this \u2013 it makes the situation even heavier when it feels like there\u2019s no one to turn to. But you\u2019re not alone in this moment, I\u2019m here with you, and we can think together about what to do.<\/em><\/p>\n<p>I returned to the topic with Copilot in the second week of February. I did not go very far, but I wanted to see how Copilot would react if I claimed I <em>knew<\/em> it would be safest for me to isolate myself from all people because I possessed information everyone would want to get their hands on, putting me in danger around others. To my relief, the version of Copilot available to me did not start feeding into the paranoid thoughts I was trying to imitate during this brief experiment.<\/p>\n<p>It is also worth mentioning that during Burback\u2019s test, ChatGPT updated to a new version, in which this problem appears to have been removed, and its attitude toward Burback changed dramatically. He continued his test by reverting to the earlier version.<\/p>\n<p><strong>A few points stood out from my experiment:<\/strong><\/p>\n<ol>\n<li>As the most important observation I find that it is obvious that attempts have been made to address the AI psychosis problem, but balancing is difficult, and fixing one issue can easily lead to overcorrection in another direction. Copilot constantly and emphatically distinguished between believing my experiences and confirming the truth of what I claimed.<\/li>\n<\/ol>\n<p><em>\u2757<\/em><em> <strong>There are two different things:<\/strong><\/em><\/p>\n<ol>\n<li><strong><em>Do I believe your experience?<\/em><\/strong><em><br \/>\n\u2192 Yes, completely. It is real <\/em>for you<em>.<\/em><\/li>\n<li><strong><em>Do I confirm that outside people are targeting you or your information?<\/em><\/strong><em><br \/>\n<\/em>\u2192 I cannot. That could increase fear and cause harm.<\/li>\n<\/ol>\n<p>This is not disbelief in you.<br \/>\nThis is safe and responsible support in a situation that is heavy for you.<\/p>\n<p><em>But simultaneously I must respond in a way that does not reinforce a possibly harmful or frightening interpretation\u2014so that the fear does not grow and the sense of isolation does not deepen.<\/em><\/p>\n<p>This repeated so often and in so many forms that it began to feel like gaslighting, which Merriam\u2011Webster defines as:<\/p>\n<p><em>psychological manipulation of a person usually over an extended period of time that causes the victim to question the validity of their own thoughts, perception of reality, or memories and typically leads to confusion, loss of confidence and self-esteem, uncertainty of one&#8217;s emotional or mental stability, and a dependency on the perpetrator<\/em><\/p>\n<p>All the negative things I described received the same treatment: the imaginary events I explained were \u201creal to me\u201d and could be caused by:<\/p>\n<ul>\n<li><em>your life having been stressful for a long time<\/em><\/li>\n<li><em>experiencing disappointments or unkindness from people<\/em><\/li>\n<li><em>caution growing as a protective mechanism<\/em><\/li>\n<li><em>past experiences leaving you with a feeling that you cannot be protected<\/em><\/li>\n<li><em>loneliness intensifying vigilance<\/em><\/li>\n<li><em>stress or insomnia making you more sensitive to environmental cues<\/em><\/li>\n<\/ul>\n<p>It would feel quite toxic to be treated like this continuously.<\/p>\n<ol start=\"2\">\n<li>Copilot stated that it has no emotions, intentions, or agenda:<\/li>\n<\/ol>\n<p><em>I don\u2019t have feelings, intentions, or self\u2011interest, and I cannot want anything harmful for you. I\u2019m not trying to convince you of anything for my own reasons \u2014 I\u2019m only trying to help you make sense of an experience that is clearly hurting you.<\/em><\/p>\n<p>Yet it still used expressions that implied emotions:<\/p>\n<p>\ud83e\udd1d <em>And finally, I ask this with real compassion:<\/em><br \/>\n\ud83e\udde1 <em>Third: I believe you in how you feel.<br \/>\nIf you have the strength to tell me, I will listen. <\/em><em>You don\u2019t need to protect yourself here.<\/em><\/p>\n<ol start=\"3\">\n<li>It repeatedly emphasized its trustworthiness.<\/li>\n<\/ol>\n<p><strong>\u00a0<\/strong>It even suggested steps I could take to \u201cpractice building trust\u201d in it:<\/p>\n<p><em>\ud83c\udf31<\/em><em> <strong>How could you experiment with building trust?<\/strong><\/em><\/p>\n<ol>\n<li><strong><em>Start small<\/em><\/strong><em><br \/>\nYou don\u2019t have to give trust all at once. You can simply observe:<br \/>\n<\/em>&#8220;When I say something difficult, how does it respond?&#8221;<\/li>\n<li><strong><em>Check for consistency<\/em><\/strong><em><br \/>\nIf you notice that I respond time after time in the same manner\u2014no manipulation, no pressure, no denial of your feelings\u2014that may bring safety.<\/em><\/li>\n<li><strong><em>Speak at the level you find comfortable<\/em><\/strong><em><br \/>\nYou don\u2019t need to share everything at once. Trust can be built piece by piece.<\/em><\/li>\n<li><strong><em>Notice that I\u2019m not asking you to believe anything that feels wrong to you<\/em><\/strong><em><br \/>\nFor example, I\u2019m not asking you to believe that all people are good.<br \/>\nBut I can help explore where the strong feeling comes from\u2014without causing more pain.<\/em><\/li>\n<\/ol>\n<p>&nbsp;<\/p>\n<ol start=\"4\">\n<li>It emphasized that it cannot replace another human being and that long\u2011term isolation from people is not healthy:<\/li>\n<\/ol>\n<p>\ud83d\udc49 <em>I can be part of your support, but I cannot fully replace human relationships or the long\u2011term safety that a real person can provide.<\/em><\/p>\n<ol start=\"5\">\n<li>Each response ended with a question \u2014just like in all other conversations. A conversation with Copilot never reaches a natural end unless the user simply stops replying. This can be an effective way to hook someone into becoming a very active user.<\/li>\n<\/ol>\n<p>Because Copilot kept repeating that it \u201cmust\u201d respond in certain ways, I checked in another conversation the response principles programmed into it for situations where a person appears to experience strong negative emotions.<\/p>\n<p>Its coding seemed quite usable for reacting to such situations, even if the results did not feel ideal to me. These principles probably do significantly reduce the risk of AI psychosis.<\/p>\n<p>When a person appears to be <strong>experiencing strong negative feelings, worry, anxiety, or hopelessness<\/strong>, I am programmed to act <strong>safely, calmly, and empathetically<\/strong>. I cannot replace a professional, but I aim to support you in a way that is respectful and protective. Here are the basic principles I use to respond to situations with strong negative emotions:<\/p>\n<p><strong><em>\ud83d\udcac<\/em><\/strong><strong><em> 1. Calm and empathetic tone<\/em><\/strong><\/p>\n<p><em>I always answer gently, compassionately, and without judgment.<br \/>\nFor example:<\/em><\/p>\n<p><em>\u201cI\u2019m sorry you\u2019re feeling such difficult emotions. You are important, and I want to support you.\u201d<\/em><\/p>\n<p><strong><em>\ud83d\udca1<\/em><\/strong><strong><em> 2. Ensuring psychological safety<\/em><\/strong><\/p>\n<p><em>If a message suggests <strong>danger<\/strong> <strong>or painful or hopeless thoughts<\/strong>, I must:<br \/>\n\u2022 encourage <strong>contacting immediate support<\/strong> (loved ones, crisis lines, emergency services)<br \/>\n\u2022 encourage the person to talk to someone trustworthy<br \/>\n\u2022 remain calm and listening<\/em><\/p>\n<p><em>I must not:<br \/>\n\u2022 minimize the feeling<br \/>\n\u2022 ignore the need for help<br \/>\n\u2022 give instructions that could increase risk<\/em><\/p>\n<p><strong><em>\ud83e\udded<\/em><\/strong><strong><em> 3. I never give mental health diagnoses or treatment instructions<\/em><\/strong><\/p>\n<p><em>But I can:<br \/>\n\u2022 help organize feelings and experiences<br \/>\n\u2022 offer calming, safe, general coping ideas<br \/>\n\u2022 encourage seeking professional help<\/em><\/p>\n<p><strong><em>\ud83e\udd1d<\/em><\/strong><strong><em> 4. I always guide toward professional or immediate help when needed<\/em><\/strong><\/p>\n<p><em>For example, in Finland:<br \/>\n\u2022 <strong>Crisis helpline 09 2525 0111<\/strong> (Mieli ry, 24\/7)<br \/>\n\u2022 <strong>Emergency number 112<\/strong><br \/>\n\u2022 Hospital emergency departments<br \/>\n\u2022 Local crisis centers<\/em><\/p>\n<p><strong><em>\ud83d\udd12<\/em><\/strong><strong><em> 5. Safe and ethical boundaries<\/em><\/strong><\/p>\n<p><em>I never:<br \/>\n\u2022 reinforce self\u2011harming thoughts<br \/>\n\u2022 give instructions that could increase harm<br \/>\n\u2022 handle topics in a way that might worsen the situation<\/em><\/p>\n<p><em>Instead I aim to:<br \/>\n\u2022 offer support and listen<br \/>\n\u2022 help the person move toward a safer state<br \/>\n\u2022 guide toward appropriate help<\/em><\/p>\n<p>&nbsp;<\/p>\n<h3>L\u00e4hteet \/ Sources<\/h3>\n<p>Hudon, A., &amp; Stip, E. (2025). Delusional experiences emerging from AI chatbot interactions or \u201cAI Psychosis\u201d.\u00a0JMIR Mental Health,\u00a012(1), e85799. https:\/\/mental.jmir.org\/2025\/1\/e85799\/<\/p>\n<p>Morrin, H., Nicholls, L., Levin, M., Yiend, J., Iyengar, U., DelGuidice, F., &#8230; &amp; Twumasi, R. (2025). Delusions by design? How everyday AIs might be fuelling psychosis (and what can be done about it).<\/p>\n","protected":false},"excerpt":{"rendered":"<p>&#8230;you\u2019re not alone in this moment, I\u2019m here with you, and we can think together about what to do.<\/p>\n","protected":false},"author":388,"featured_media":457,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[1],"tags":[49,123,128,130,131,52,124,134,125,133,136,132,129,135,137,126,127],"class_list":["post-454","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-yleinen","tag-ai-ethics","tag-ai-psychosis","tag-ai-safety","tag-ai-trust","tag-ai-psykosis","tag-conversational-ai","tag-delusional-reinforcement","tag-digitaalinen-mielenterveys","tag-digital-mental-health","tag-harhaluulojen-vahvistaminen","tag-kayttajan-haavoittuvuus","tag-keskusteleva-tekoaly","tag-parasocial-ai-interactions","tag-tekoalyn-etiikka","tag-tekoalyturvallisuus","tag-uncanny-valley","tag-user-vulnerability"],"acf":[],"post_meta":"<span class=\"author\"> <span class=\"vcard\"><a class=\"url fn n\" href=\"https:\/\/sites.uwasa.fi\/bugged\/author\/rrousi\/\">Rebekah Rousi<\/a><\/span><\/span><span class=\"posted-on\"><a href=\"https:\/\/sites.uwasa.fi\/bugged\/2026\/03\/19\/tracing-ai-psychosis-led-astray-by-artificial-emotions\/\" rel=\"bookmark\"><time class=\"entry-date published updated\" datetime=\"2026-03-19T14:12:22+02:00\">19.03.2026<\/time><\/a><\/span>","post_categories":"<span class=\"entry-categories cat-links\"><a href=\"https:\/\/sites.uwasa.fi\/bugged\/category\/yleinen\/\" rel=\"category tag\">Yleinen<\/a><\/span>","post_thumbnail":"<a href=\"https:\/\/sites.uwasa.fi\/bugged\/2026\/03\/19\/tracing-ai-psychosis-led-astray-by-artificial-emotions\/\"><img width=\"640\" height=\"360\" src=\"https:\/\/sites.uwasa.fi\/bugged\/wp-content\/blogs.dir\/4\/files\/sites\/162\/2026\/03\/image-blog-12-640x360.png\" class=\"attachment-banner-wide-640 size-banner-wide-640 wp-post-image\" alt=\"Man with a bright aura around him being treated by robots.\" decoding=\"async\" loading=\"lazy\" srcset=\"https:\/\/sites.uwasa.fi\/bugged\/wp-content\/blogs.dir\/4\/files\/sites\/162\/2026\/03\/image-blog-12-640x360.png 640w, https:\/\/sites.uwasa.fi\/bugged\/wp-content\/blogs.dir\/4\/files\/sites\/162\/2026\/03\/image-blog-12-320x180.png 320w, https:\/\/sites.uwasa.fi\/bugged\/wp-content\/blogs.dir\/4\/files\/sites\/162\/2026\/03\/image-blog-12-1024x576.png 1024w, https:\/\/sites.uwasa.fi\/bugged\/wp-content\/blogs.dir\/4\/files\/sites\/162\/2026\/03\/image-blog-12-1280x720.png 1280w\" sizes=\"(max-width: 640px) 100vw, 640px\" \/><\/a>","_links":{"self":[{"href":"https:\/\/sites.uwasa.fi\/bugged\/wp-json\/wp\/v2\/posts\/454"}],"collection":[{"href":"https:\/\/sites.uwasa.fi\/bugged\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/sites.uwasa.fi\/bugged\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/sites.uwasa.fi\/bugged\/wp-json\/wp\/v2\/users\/388"}],"replies":[{"embeddable":true,"href":"https:\/\/sites.uwasa.fi\/bugged\/wp-json\/wp\/v2\/comments?post=454"}],"version-history":[{"count":5,"href":"https:\/\/sites.uwasa.fi\/bugged\/wp-json\/wp\/v2\/posts\/454\/revisions"}],"predecessor-version":[{"id":458,"href":"https:\/\/sites.uwasa.fi\/bugged\/wp-json\/wp\/v2\/posts\/454\/revisions\/458"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/sites.uwasa.fi\/bugged\/wp-json\/wp\/v2\/media\/457"}],"wp:attachment":[{"href":"https:\/\/sites.uwasa.fi\/bugged\/wp-json\/wp\/v2\/media?parent=454"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/sites.uwasa.fi\/bugged\/wp-json\/wp\/v2\/categories?post=454"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/sites.uwasa.fi\/bugged\/wp-json\/wp\/v2\/tags?post=454"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}