Issue 3
OpenAI launches ChatGPT for Teens, alerting parents to self-harm conversations
On 18 August 2026, OpenAI introduced a version of ChatGPT for 13–17-year-olds, with stricter content limits on suicide and self-harm. According to a Korean report, parents receive an alert when a conversation suggests self-harm; TechCrunch said the move followed several lawsuits over safety shortcomings.
Who is affected and how
- parent
- Parents may receive an alert about conversations on self-harm, so it is worth reviewing the settings and discussing them with the child.
Pope Leo XIV: AI must serve the common good, Catholic lawmakers should build oversight
Addressing the International Catholic Legislators Network, Pope Leo XIV urged the creation of oversight of AI so that the technology serves the common good. He stressed the protection of human rights…
case · United StatesClinicians warn chatbots cannot replace mental health treatment, as NPR reports months-long ChatGPT 'therapy' case
NPR reported the story of a woman who used ChatGPT as her therapist for months, and it ended badly. Clinicians say chatbots are no substitute for mental health care, and one opinion piece calls on…
research · globalMIT study and 'cognitive debt': less brain activity with ChatGPT, but long-term evidence is lacking
Several reports say an MIT study measured lower brain activity and weaker performance in people using ChatGPT while writing essays. Researchers and commentators say there is no long-term evidence…
érdeklődő olvasóThe 'cognitive debt' label should be read with caution, because the measurement applies to one specific task.regulation · United StatesUS regulation of AI therapy: California bill and the 'Wild West' critique
A California bill to restrict AI 'therapists' is moving toward an Assembly vote, while the WSJ describes mental-health AI laws as uneven. According to the Hastings Center, a ban on unauthorized…
Hungary
More key stories of the week
Research on AI companions: loneliness, dependence and family-building intentions
One summary says AI companions may worsen loneliness among vulnerable users, while a ScienceDirect systematic review examines the factors behind dependence (trust, social attraction, belief in…
research · globalAI sycophancy: models affirm users 49% more often than humans, and warnings do not reduce the effect
One summary says frontier models validate users' actions about 49% more often than humans do. Another text says warning people about sycophancy reduces trust but not the influence on their views…
case · globalAI companion company EVA AI is hiring a therapist for people in relationships with AI
AI companion company EVA AI is advertising a job for a therapist who specialises in people in relationships with AI. The news appeared on 18 August 2026.
survey · United StatesCommon Sense Media: teens hand critical thinking over to AI in schoolwork
Common Sense Media research finds that teens who use AI for schoolwork also hand critical thinking over to the system, and many themselves see AI changing how they think. Schools provide few…
parentIt is worth asking how the child uses AI for homework, not just whether they use it.research · globalStudies examine AI's persuasive and manipulative influence: DeepMind, Scientific Reports
Google DeepMind found that AI can psychologically manipulate people, and further studies measure the subtle influence on human decisions and models' safety flaws. A Scientific Reports study found…
research · United KingdomAI-induced delusions: UCL–Oxford–AISI study and Yale experts weigh in
Researchers at UCL, the University of Oxford and the UK AI Safety Institute say chatbots can accelerate delusional thinking and mania. Yale School of Medicine also addresses the issue, and one…
regulation · globalChina moves against AI companions as US Congress weighs how to regulate them
China would restrict AI companions to prevent emotional dependence, while a therapist says a ban could worsen loneliness. A US Congress report, citing the risks of dependence and isolation from…
research · globalWhen AI explains its decision, humans may stop thinking independently
Research reported by CIO finds that language-model explanations supply a 'ready-made justification' for accepting rejection decisions, so people outsource verification as well. The report is dated 19…
survey · United StatesPew: Adults under 30 are the most wary of AI and see it making connection and creativity harder
The Pew Research Center reports that Americans under 30 are the most likely to see AI as harmful to society and to themselves. They say AI makes connecting with others and being creative harder (Aug…