Last Month in AI x Humanity

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:

• Anthropic study links Claude’s reasoning to global workspace theory

• FTC seeks public comment on a bias-related AI policy statement

• AI labs continue to hire more philosophers for reasoning and ethics

• Major AI researchers leave Google for OpenAI and Anthropic

• UN report cautions about AI’s broader environmental impact

Anthropic researchers published a study arguing that Claude has developed something like an internal “workspace” for reasoning, drawing on global workspace theory, an idea that consciousness works a bit like a theater stage where most processing happens in the background, but some information gets selected and shared more widely across the brain. The researchers call Claude’s version “J-space,” and say it helps the model hold ideas in mind, reason silently, and report what it was thinking about. They emphasize that this does not show Claude is conscious, but it may give researchers a clearer way to study what advanced AI systems are processing internally.

The Federal Trade Commission is asking for public comment on a proposed policy statement about whether AI companies may be shaping their systems’ answers in ways users would not expect. The proposal says companies could violate consumer protection law if they present AI tools as accurate or objective while secretly steering outputs toward undisclosed goals. The FTC is also weighing how federal rules may interact with state AI laws, with public comments open through July 31, 2026.

The Economist reports that major AI labs are increasingly hiring philosophers to help shape how AI systems reason and handle ethical issues. The idea is that philosophy training can be useful for work on logic, safety, and decision-making as models become more powerful. Some philosophers are reportedly helping write company principles and model “constitutions,” raising broader questions about who gets to decide what values AI systems should follow.

A new BCG survey finds that AI use at work is spreading quickly, especially among frontline employees, with many regular users saying it saves them about a full workday each week. But the report also says many companies have not clearly explained how workers should use that saved time or redesigned workflows around AI. BCG argues that the main challenge is shifting from simply giving employees AI tools to creating clearer plans, training, and governance for how people and AI systems should work together.

OpenAI released a report on how its employees and customers are using Codex, its AI coding agent, as work shifts from short chatbot exchanges to longer delegated tasks. The report says Codex is now widely used across OpenAI departments, including legal, finance, and recruiting, not just engineering. OpenAI also found that non-developers are one of the fastest-growing user groups, suggesting that AI agents may increasingly help people do technical or analytical work outside their usual job roles.

Several major AI researchers have reportedly recently moved between top labs, including Noam Shazeer leaving Google for OpenAI, John Jumper leaving Google DeepMind for Anthropic, and Andrej Karpathy joining Anthropic. The moves show how intensely AI companies are competing for a small group of highly influential researchers, especially as coding and advanced reasoning become central areas of competition. The departures are also notable for Google, which has recently regained momentum in AI but continues to face pressure from OpenAI and Anthropic.

A new Pew Research Center survey finds that about half of U.S. adults now use AI chatbots, up from about a third in 2024, with many using them for search, work tasks, and everyday questions. At the same time, Americans remain cautious about AI’s broader impact: most say the technology is advancing too quickly, and many worry it will make personal information less secure. The survey also finds skepticism across both parties, though Democrats are now somewhat more likely than Republicans to lack confidence in the government’s ability to regulate AI effectively.

Anthropic released a study analyzing about 400,000 Claude Code sessions to understand how people are using AI coding agents in practice. The report finds that users usually decide what they want done, while Claude handles much of how to do it, such as editing code, running commands, or fixing issues. It also suggests that success depends less on whether someone is a professional programmer and more on whether they understand the problem they are trying to solve, pointing to a future where coding tools may help more non-programmers do technical work.

A new United Nations University report warns that AI’s environmental impact is broader than just carbon emissions from data centers. The report says AI systems also use significant amounts of water and land through electricity generation, cooling, hardware supply chains, and electronic waste. It argues that as everyday AI use grows, especially image and video generation, policymakers and companies may need to measure AI’s full footprint more clearly and consider where data centers are built, how much energy they use, and who bears the local costs.

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.