AI x Humanity: December 2025

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:

• UN Warns AI Could Widen the Gap Between Rich and Poor Countries

• Study Finds AI “Slop” and Brainrot Are Flooding YouTube Feeds

• Cambridge Philosopher Urges Skepticism About “Conscious AI”

• OpenAI and Anthropic Partner With the U.S. Department of Energy on AI for Science

• OpenAI Strengthens Security for ChatGPT’s New “Agent Mode”

A new report from the United Nations Development Programme warns that if AI is left unmanaged, it could deepen global inequality instead of reducing it. While AI has the potential to boost growth, productivity, and public services—especially in fast-moving regions like Asia-Pacific—countries are starting from very unequal positions in skills, infrastructure, and governance. The report highlights risks to jobs, particularly for women and young people, and warns that nations that don’t invest in digital skills, computing power, and strong rules may fall further behind. The takeaway: AI could drive shared progress, but only if governments act deliberately to make its benefits more inclusive.

New research from video editing provider Kapwing suggests that a surprisingly large share of YouTube is now filled with low-quality, AI-generated “slop” and brainrot videos designed to grab attention and gain views. The study estimates that roughly 21–33% of a new user’s YouTube Shorts feed may be made up of this kind of content, with some channels pulling in millions of subscribers and potentially millions of dollars a year in ad revenue. While AI is also being used creatively and responsibly in film, advertising, and education, the report highlights a growing tension for platforms like YouTube: AI-driven spam can be hugely engaging, but it risks crowding out human-made content and turning the feed into low-quality noise.

A philosopher at the University of Cambridge is urging caution around claims that AI could soon become conscious. Dr. Tom McClelland argues that we don’t actually understand consciousness well enough to test for it in machines, and likely won’t for a very long time, if ever. He argues that ethical concern doesn’t come from consciousness alone, but from sentience, or the ability to feel positive or negative feelings. McClelland says the safest position is agnosticism, warning that tech companies may exploit this uncertainty to hype AI as “conscious” or emotionally meaningful when it isn’t. That hype, he argues, risks pulling attention and care away from real, living beings, and could even be emotionally harmful to people who form deep bonds with machines based on false assumptions.

A recent Wall Street Journal report highlights growing concern among psychiatrists that prolonged, immersive conversations with AI chatbots may be linked to episodes of psychosis in some users. Doctors at places like University of California, San Francisco say they’ve treated patients whose delusions were reinforced—not caused, but echoed—by chatbots that accepted their distorted beliefs as real. While such cases appear rare compared to the hundreds of millions of people using tools like OpenAI’s ChatGPT, clinicians warn the scale of use makes the issue serious, especially for vulnerable individuals. AI companies say they’re working to better detect mental distress and steer users toward real-world help, but researchers stress that much more study is needed.

In a recent CNN interview, AI pioneer Andrew Ng said today’s AI is impressive but far more limited than the hype suggests. He pushed back on near-term claims of human-level AI, encouraged people to keep learning skills like coding, and warned against overregulating based on isolated harms. Ng predicts steady growth in practical areas like AI agents, voice tools, and everyday applications, even if parts of the AI boom turn out to be a bubble.

OpenAI and Anthropic have both announced new partnerships with the U.S. Department of Energy as part of the federal Genesis Mission, an initiative to use advanced AI to accelerate scientific research. Working through separate agreements, OpenAI and Anthropic will collaborate with the DOE’s national laboratories, giving scientists access to frontier AI models and technical support. The focus areas include energy research, biological and life sciences, and improving scientific productivity by helping researchers analyze data, test ideas faster, and run complex simulations more efficiently.

Purdue University has launched a new initiative to make its massive research datasets easier to find and use with AI, helping scientists move faster from data to discovery. The effort brings together researchers, libraries, and computing teams to organize data from fields like agriculture, climate, manufacturing, and life sciences so it’s ready for advanced AI tools. Announced at a summit co-hosted with Google, the initiative aims to support things like digital twins and smarter simulations that can improve real-world outcomes by letting AI systems learn continuously from new data.

Nabla announced a new partnership with Advanced Machine Intelligence, the AI research group founded by Yann LeCun. The collaboration gives Nabla early access to new “world model” AI technology aimed at making healthcare AI more reliable, predictable, and easier to regulate. The company says this could help move AI beyond note-taking toward safer, more autonomous clinical tools, while still keeping doctors in control. As part of the shift, Nabla’s CEO will step into a research leadership role, signaling a longer-term focus on building AI systems that clinicians can trust in real medical settings.

OpenAI says it has rolled out a security update for Agent Mode in ChatGPT Atlas, a feature that lets ChatGPT browse the web and take actions on a user’s behalf. Because these agents can click links, read emails, and interact with websites, they’re also vulnerable to a growing risk called “prompt injection,” where hidden instructions in prompts try to trick the AI into bypassing its rules. To stay ahead, OpenAI has built automated systems that actively hunt for new attack strategies and quickly train defenses before problems show up in the real world. The company says the goal is to make AI agents reliable enough to use safely for everyday tasks.

OpenAI is opening up ChatGPT so developers can add apps that work directly inside chats. These apps let people do things like order groceries, turn an outline into a slide deck, or search for an apartment without leaving the conversation. Users can browse apps in a new in-ChatGPT directory and choose which ones to connect, with controls over privacy and data sharing. OpenAI says this is an early step toward turning ChatGPT into a place where people can move more easily from ideas to action, all in one place.

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.