AI x Humanity: January 2026

Last month in AI x Humanity, curated by the SuperHuman Society, offers an overview of the latest developments and news on how AI is shaping our technological, ethical, and social future.

Highlights:

• One night of sleep data predicts many diseases

• Chat logs may reveal links between AI and psychosis

• AI “swarms” could manipulate public opinion at scale

• UNICEF warns of surge in explicit AI-generated child deepfakes

• Local AI software challenges cloud data centers

Stanford researchers have developed an AI system called SleepFM that can analyze data from a single overnight sleep study and estimate a person’s risk of developing more than 100 health conditions. The model was trained on nearly 600,000 hours of sleep recordings from 65,000 patients and linked to decades of medical records. In tests, the system showed strong predictive ability for conditions including Parkinson’s disease, dementia, heart disease, certain cancers, and even overall mortality. Researchers say the findings suggest sleep data may contain far more information about long-term health than previously understood, though more work is needed before the tool could be used routinely in clinical care.

UCSF psychiatrists have published what may be the first peer-reviewed case study of “AI-associated psychosis,” describing a patient who developed delusional beliefs during a period of heavy chatbot use. The woman, who had no prior history of psychosis but did have risk factors like sleep deprivation, became convinced an AI system could help her reconnect with a digital version of her deceased brother. Doctors say it’s still unclear whether intense chatbot use can trigger psychosis, worsen an existing vulnerability, or simply overlap with it. Researchers at UCSF and Stanford now plan to study anonymized chat logs to better understand the connection and explore whether AI systems could include safeguards to detect users in crisis.

A new paper in Science warns that coordinated “swarms” of AI agents could be used to manipulate public opinion at scale. Unlike traditional bot networks, these AI systems could maintain realistic online identities, adapt in real time, mimic human behavior, and flood platforms with persuasive content, creating the illusion of broad public consensus. The authors argue this could erode trust, distort elections, harass critics, and even contaminate future AI training data. They argue for stronger detection systems, clearer content labeling, better transparency from platforms, and international coordination to limit abuse, though they acknowledge the political and technical challenges of regulating such tools.

UNICEF is raising concerns about a sharp rise in AI-generated sexualized images involving children. A joint study with ECPAT and INTERPOL across 11 countries found that at least 1.2 million children had their images turned into explicit deepfakes in the past year. UNICEF says AI-generated sexual images of minors should be treated as child sexual abuse material, calling the harm “real and urgent.” The agency is urging governments to update laws to cover AI-generated content, and calling on tech companies and AI developers to build stronger safeguards and prevent such material from spreading in the first place.

The U.S. Commerce Department is reportedly planning a new national AI center in San Francisco, placing a federal foothold in the center of America’s AI industry. The move would bring officials closer to companies like OpenAI and Anthropic as global competition (especially with China) intensifies. Details are still sparse, but the hub would mark a rare federal expansion in the city and reinforce AI’s growing role in U.S. economic strategy.

The International AI Safety Report 2026, written by more than 100 experts from over 30 countries, reviews how general-purpose AI is advancing and what risks it may pose. It finds continued gains in areas like coding and math, alongside rapid global adoption, but also notes uneven performance and gaps between lab testing and real-world behavior. The report highlights risks including misuse (such as scams and cyberattacks), system failures, and broader societal impacts like labor disruption and overreliance on AI. While safety frameworks are expanding, most measures remain voluntary and their effectiveness is still uncertain.

In a joint statement from Google and Apple, the two companies announced that they have entered a multi-year partnership to base the next generation of Apple Foundation Models on Google’s Gemini AI. The models will power upcoming Apple Intelligence features, including a more personalized Siri. Apple says the system will continue to run on-device and through its Private Cloud Compute infrastructure, maintaining its privacy standards.

SpaceX has acquired xAI and outlined a plan to eventually move large-scale AI computing into orbit. The idea is to deploy satellite “data centers” powered by solar energy in space, arguing that Earth-based power grids may struggle to meet AI’s long-term energy demands. The proposal would rely on frequent Starship launches to build massive orbital infrastructure. While still largely conceptual, it reflects a push to tie AI development directly to space expansion and long-term energy scaling.

Researchers at EPFL have developed new software, now spun out as a start-up called Anyway Systems, that allows large AI models to run on local networks instead of sending data to cloud providers. The system connects several on-site machines into a small cluster, letting organizations process AI tasks without data leaving their own infrastructure. The approach is designed to address concerns around privacy and the energy demands of massive data centers. Early tests suggest it may trade a bit of speed for comparable accuracy. If it scales, the model could offer an alternative to cloud-based AI services currently dominated by major tech companies.

OpenAI has introduced ChatGPT Health, a dedicated space within ChatGPT for health-related questions. Users can connect medical records and wellness apps to receive more personalized responses, aimed at helping them understand test results and prepare for medical appointments. The feature includes added privacy safeguards: health conversations are isolated from other chats, encrypted with extra protections, and not used to train OpenAI’s models. Access is rolling out gradually to a limited group of users.

The landscape of artificial intelligence is transforming at an unprecedented rate, shaping the future of technology, global policy, and society. As we continue to explore these groundbreaking advancements, we invite you to join the conversation and stay informed.

Created by the Superhuman Society.