GUEST BLOG: Talk Liberation – AI Everything: Chatbots To Be Your Friend, Your Doctor and Your Judge – What Could Go Wrong?
We’re delivering you the hottest internet news that affects all of us. Scroll down to read our full reporting below and if you love what we’re doing, please consider contributing $5 per month so that we can continue providing you with this vital, unique digest.
In this edition:
- OpenAI + MIT Study Reveals Loneliness Caused By Chatbot Usage
- Colorado Proposes New Framework for Responsible AI Policy
- AI Already In Use By Judges Deciding Cases
- Graphene OS Commits To Non-Compliance With Age Verification Laws
- US Defence Contractor Palantir Accesses UK NHS Healthcare Data
- US FCC bans all foreign manufactured routers
- AI Authorised To Prescribe Mental Health Drugs In Utah
- AI Found To Misdiagnose Up To 80% Of Medical Cases
OpenAI + MIT Study Reveals Loneliness Caused By Chatbot Usage
A paper written by multiple MIT graduates outlines the psychological consequences of integrating chatbots into daily life. Authors of the paper include Cathy Mengying Fang, an MIT graduate student who has received a Best Paper Award as well as multiple honorable mentions for her work on studying how AI systems affect human lives, and Pattie Maes, an award-winning and long time professor at MIT.
The study included 981 people and over 300,000 messages, focusing on people who are reported to have ‘moderate’ levels of loneliness and socialisation with other humans that are similar to what is average among the larger population.
The report goes on to explain that the participants were advised to take 5 minutes a day talking to the chatbot, and the actual numbers of minutes were a minimum of 1.01 and a max of 27.65.
When closely examining the differences in effects between those who spent less vs more time with the bot the researchers noticed a trend that predicted higher loneliness, less socialisation, and more problematic use of the chatbot.
Colorado Proposes New Framework for Responsible AI Policy
The US state of Colorado proposed a bill for an act “concerning the use of automated decision making technology in consequential decisions” in March 2026. The proposed legislation would force insurance companies to make known to its clients whenever AI is used in consequential decision-making regarding their applications.
This bill has been described by Governer Jared Polis as a framework for Colorado’s future AI policies “to protect consumers and support innovation in our state” after an initial unanimous agreement by Colorado’s AI Policy Working Group.
Fastforward two months later to May 2026, and a “watered down” version of the Bill has been approved by State legislators and sent to the office of the Governor for signing. The revised bill pushes the date for implementation out by an additional six months to January of 2027, retaining and enhancing consumer protections but easing the imposition of some direct restrictions upon companies utilising AI. State representative Javier Mabrey is quoted by the Colorado Sun as saying he voted for the bill to pass because “some regulation is better than none.”
AI Already In Use By Judges Deciding Cases
A study published by Northwestern University, co-published by the New York City Bar, included surveys on US judges about their opinions about, but most importantly their usage of, Artificially Generated Content in regards to their work life.
Staggeringly, over 60% of US judges confessed to having used at least one AI tool for their judicial work. 22.4% said they used those tools weekly or daily – making nearly one in four.
There were 502 judges sampled in the study, including judges for bankruptcy, Magistrate’s, Court of Appeals, and District Court. The study did find judges were more likely to use AI marketed as being for law, such as CoCounsel, Lexis+ AI, or Protégé.
Those judges reported their usage of AI going towards legal research and reviewing documents, and when asked about others working in their chambers the percentage slightly increased (From 30% to 39.8% and 15.5% to 16.7% respectively.)
Given AI’s widely reported penchant for hallucination and/or the purveying of misinformation, reliance on these tools or their potential to influence judicial decision-making marks a worrying turn.
Graphene OS Commits To Non-Compliance With Age Verification Laws
Graphene OS, an android operating system focused on privacy and the protection of users, is taking a public stance refusing to comply with new bills (such as Brazil’s Digital Statute for Children and Adolescents, just as one of many examples) which requires people to verify their age before using an operating system.
Graphene have stated they will continue to make their system available and privacy-respecting internationally even if they are banned by certain countries. When someone asked about them potentially blocking VPN’s access to their site, they responded with “We don’t filter the internet for Iran or North Korea so why would we for Brazil or California?”
This month, Graphene released a lengthy statement alleging anti-competitive practices by both Google and Apple. The allegations revolve around the Big Tech giants creating service dependencies that make it impossible for them to be cut out of the loop by competing operating systems offering meaningful alternatives to consumers.
US Defence Contractor Palantir Accesses UK NHS Healthcare Data
Palantir Technologies, a notorious AI-centric data analytics company servicing governments, defence contractors and intelligence agencies around the world, secured a £330 million NHS contract with England – gaining it access to England’s main platform for medical data shared between hospitals.
In response, the NHS directly received many pleas from human rights, health and patient unions and organisations, including 50,000 patients who sent emails to trust boards in their local hospitals asking them not to integrate their systems with or provide access to Palantir. It has been asserted that hospitals have the ability to outright refuse this new directive, and many have urged them to do so.
There is additionally growing concern from the public due to Palantir also being used by ICE in the US to track people down, including using their health data.
Despite the pleas of the general public, this month the NHS confirmed that Palantir staff are indeed being allowed to access NHS patients health data.
US FCC bans all foreign manufactured routers
The US FCC has moved to ban all imports of new foreign-made consumer routers. Even though the move is for all foreign-made routers, estimates show China controls at least 60% of the market for routers and US lawmakers have previously raised concerns about Chinese-made routers.
John Moolenaar, the Republican chair of the House select committee on China, praised the FCC order, stating; “Routers are key to keeping us all connected and we cannot allow Chinese technology to be at the center of that.”
AI Authorised To Prescribe Mental Health Drugs In Utah
Legion health, a new AI startup with 7 million in funding, has been given the power to prescribe certain anti-depressants and other psych drugs. The rollout requires only the first 1,250 prescriptions to have any human oversight at all, before the AI is left to prescribe these medications completely on its own.
Legion’s plan is for AI to run refills on already prescribed medications, especially for Utah whose more rural areas sometimes have no practicing psychiatrists.
The AI-prescriptions are pushing ahead despite researchers having proven the AI can be manipulated to produce dangerous outcomes. It was evidenced that the Utah prescription AI had been “tricked” into increasing an Oxytocin prescription by three times as well as getting the AI to encourage vaccine conspiracies, and to suggest methamphetamine as a treatment. Axios reported: “Critics warned this pilot could create safety risks — and researchers say the flaws persist, despite alerting the company in January.”
AI Found To Misdiagnose Up To 80% Of Medical Cases
A study published by Jama Network Open, including assessments from 21 language models relating to 29 standardised medical case scenarios with over 16,000 responses generated, found failure in over 80 percent of early-stage cases. In many cases where the AI was found to be correct, the diagnosis was not thought to be the most likely which diminishes the practical value even more.
In scenarios with a final diagnosis, where the problem had surpassed early stages and symptoms were more obvious, the failure rate dropped to below 40% however in the world of diagnosis early detection can be the most critical, leading to skepticism about AI’s utility for real world scenarios.
