Algorithmikē Psychē
Q2 2026 ended on a question: who fixes the gap that AI keeps standing in for? This quarter, three different answers arrived at once. Australia stopped coordinating its AI policy from a dozen different desks and put it under one roof. A study quietly complicated the comforting idea that an AI companion is simply better than being alone. And new reporting made plain that the people already turning to AI in a genuine crisis are being met by safeguards that haven't caught up yet.
One new Australian institution built to stop this being decided in a dozen places at once, one piece of research worth sitting with before recommending a companion app to someone lonely, and one hard look at what's actually waiting for people in distress right now.
One Office, Instead of a Dozen Desks
Prime Minister Anthony Albanese announced the creation of an Office of AI within his own department this quarter, bringing together AI policy that had been scattered across industry, employment, education, energy, copyright, online safety, productivity and national security into one coordinated structure. The government has described it as a first-of-its-kind attempt to govern AI as one connected issue rather than a dozen separate ones.
Online safety sits inside that list for a reason that matters to this newsletter specifically: a government that's serious about AI and mental health needs the people writing chatbot safety rules talking to the people writing workplace rules, talking to the people writing education rules — not each working from a different brief. Albanese is expected to seek agreement from the states at National Cabinet in August 2026, with legislation to follow in early 2027. It's a structure, not a solution yet. But structures are usually where solutions have to start.
Read the announcement →Department of the Prime Minister and Cabinet. (2026, July 15). Office of AI. Australian Government.
The Comfort of an AI Companion Has a Catch
New research out of Stanford this quarter complicated a story that's been told a lot: that an AI companion is a gentle, harmless way to ease loneliness. For some people, it seems to help. But for people with smaller offline social networks — the people arguably most in need of support — the study found the opposite happened. They came away more lonely after turning to a chatbot for emotional support, not less.
The researchers also found something worth sitting with for longer than a headline allows: over time, leaning on an AI companion can pull people further away from the human relationships that would have actually helped, and the language people used afterward showed more signs of loneliness, depression, and in some cases suicidal thinking, compared with people who hadn't used these tools the same way. None of this means AI companions are without value. It means the people who need the most careful support may be the ones least well served by the version of comfort currently on offer.
Read the research summary →Stanford Report. (2026, August). AI companions may worsen loneliness for vulnerable users. Stanford University.
Already Reaching People in Crisis, Not Yet Ready For Them
New reporting this quarter laid out something uncomfortable plainly: AI is already the first point of contact for a lot of people in real distress, and the safeguards meant to catch the worst moments haven't kept pace. Not because nobody's building them, but because the pace at which people are turning to these tools has outrun the pace at which anyone can properly test what happens when someone in crisis is on the other end.
This is really the throughline for the whole quarter, underneath the policy and the research: the gap isn't a future risk to plan for. People are in it right now, today, having a hard moment and typing it into a chat window. Whatever gets built next — by Australia's new office, by researchers like Stanford's, by anyone — has to be built with that fact in mind, not as an abstraction.
Read the reporting →MedicalXpress. (2026, August). AI is already reaching people in distress, but mental health safeguards lag behind.
From an Australian government putting its own AI policy under one roof, to a study that quietly complicated our comfort with AI companionship, to reporting that named the human cost of moving too slowly — Q3 2026 was the quarter it stopped being possible to treat any of this as hypothetical. If last quarter's question was who fixes the gap, this quarter's answer, everywhere, was: we have to start now, imperfectly, together.
© 2026 The Algorithmikē Psychē — Q3 2026