Last Month in AIxHumanity

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:

• U.S. Government Blocks Access to Anthropic’s Advanced Models

• U.S. Officials Discuss Taking Stakes in AI Companies

• SpaceX Goes Public

• The Debate Over Whether AI is Conscious Continues Among High Profile Individuals

Anthropic said it disabled access to its Fable 5 and Mythos 5 AI models after a U.S. government order blocked access by foreign nationals, including some of its own employees. The models were part of Anthropic’s latest group of more powerful releases, with particular concern around cybersecurity uses. The government cited national security concerns about a possible workaround that could bypass the models’ safeguards. Anthropic said it disagrees with the decision, but disabled the models for all users because it could not easily limit access only to U.S. citizens. Reuters reported that the move marks a shift from regulating AI chips toward restricting access to advanced AI models themselves.

The Wall Street Journal reports that senior U.S. officials have discussed the idea of the federal government taking equity stakes in major AI companies, after OpenAI CEO Sam Altman raised a related proposal with the administration last year. Supporters argue this could let the public share in the financial upside of AI and give companies a clearer relationship with regulators, while critics may worry about market risk and closer government involvement in private AI firms. The discussions come as several AI companies move toward public markets and as policymakers debate how to respond to AI’s potential effects on jobs, wealth, and economic power.

OpenAI announced Rosalind Biodefense, a new program that gives selected public health, research, and government partners access to GPT-Rosalind, an AI model built for life sciences work. The program is focused on areas like disease detection, outbreak planning, vaccine and treatment research, and biological threat screening. OpenAI says the goal is to help trusted organizations use advanced AI for public health and biodefense while keeping additional safeguards in place around sensitive biological uses.

A Vanderbilt University report analyzes leaked documents from Geedge Networks, a Chinese company tied to internet monitoring and censorship tools. The report says Geedge was exploring ways to use AI to combine data like internet activity, location history, and social connections into profiles that could help predict who might be seen as politically risky. The researchers also argue that U.S. chip restrictions appear to have slowed some of this work by limiting access to the computing power needed to build more advanced surveillance systems.

SpaceX began trading publicly after pricing its initial public offering at $135 per share, with shares opening higher and closing its first day at $161.11. The listing valued the company at more than $2 trillion and made it the largest IPO on record. The New York Times reports that the debut also made Elon Musk the world’s first trillionaire and could help set the stage for other large AI-focused companies, including OpenAI and Anthropic, to pursue public listings.

DeepSeek released a preview of DeepSeek-V4, its newest AI model family, with two versions: a larger Pro model and a faster, cheaper Flash model. The company provided an overview of V4, discussing how it’s designed to perform strongly on reasoning, coding, and longer tasks, while supporting much larger amounts of text at once than earlier models. DeepSeek is also positioning the release as an open model alternative to leading closed AI systems, with access available through its chat app and API.

AI pioneer Geoffrey Hinton said in a recent Big Technology Podcast interview that he believes today’s AI systems may already be conscious, arguing that their ability to understand language, respond flexibly, and recognize when they are being tested suggests something more than simple pattern-matching. However, science-fiction writer Ted Chiang has argued quite the opposite, saying that fluent chatbot responses can make AI seem like a thinking being even when it is really producing convincing text. The debate shows how quickly AI is raising questions about how people understand intelligence, consciousness, and human uniqueness.

Microsoft researchers introduced SocialReasoning-Bench, a new test for whether AI agents can represent a user’s interests in everyday social tasks like scheduling meetings or negotiating purchases. The study finds that current AI models often complete the basic task, but do not always get the best outcome for the user, such as accepting worse meeting times or less favorable deals. The researchers argue that as AI agents take on more real-world responsibilities, developers need better ways to measure whether they are acting carefully and reliably on a person’s behalf.

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.