Anthropic announced in April that its Claude Mythos model had found vulnerabilities in every major operating system and web browser, and declared it too dangerous for public release. The company restricted access to about 50 trusted organizations under Project Glasswing, giving defenders a head start to patch flaws before attackers could exploit them. OpenAI followed a week later with a limited release of its cybersecurity-focused GPT-5.4-Cyber. Experts say this marks a turning point: AI labs may increasingly withhold their most capable models, treating them as dual-use technology. The move reignites the debate over open versus closed AI. Some researchers warn that if governments classify powerful AI as defense-relevant, extra controls could limit who gets to use it. Unauthorized access to Mythos has already occurred, raising questions about how long restricted releases can remain contained.
Anthropic, based in San Francisco, is a leading AI safety-focused company founded by former OpenAI employees. Its Claude models compete with OpenAI's GPT series. The restricted release aims to let cybersecurity defenders patch vulnerabilities before attackers can use them, but critics argue it concentrates power and reduces transparency.
The Verge traces the deep entanglement of AI and the military, from the 2017 Project Maven controversy—when Google helped the Pentagon analyze drone footage—to Anthropic's current fight to preserve 'red lines' banning fully autonomous lethal weapons and domestic mass surveillance. Anthropic is the only military AI contractor to impose such limits, but the article argues that AI has been transforming warfare for decades, enabling faster and more precise killings. At a 2017 UN meeting, a short film called Slaughterbots shocked attendees by showing how close autonomous killer drones had become. The piece suggests that Anthropic's red lines may not hold, given the Pentagon's long-standing investment in AI and the lack of international agreements. The history shows that once a capability is developed, restraint is rare.
Project Maven was a Pentagon initiative using AI to analyze surveillance video; Google's involvement sparked employee protests and a company pledge not to develop AI for weapons. Anthropic, known for its safety research, is now a major defense contractor. The Convention on Certain Conventional Weapons has discussed lethal autonomous systems since 2014 but has not produced a binding treaty.
Chinese authorities executed Xu Yao on May 21 for the 2020 murder of Lin Qi, the 39-year-old billionaire founder of Yoozoo Games. Xu had been sidelined by Lin after helping secure a Netflix deal for the adaptation of Liu Cixin's 3 Body Problem trilogy. He disguised lethal substances as probiotic pills and gave them to Lin, who died nine days later. Several other people also fell ill from the poisoning. A Shanghai court convicted Xu in 2024, calling his plot 'extremely despicable.' Lin's company held the film rights to the sci-fi series; the Netflix show became one of the platform's most-watched, with Lin credited posthumously as executive producer. Yoozoo Games confirmed the execution, stating 'justice has ultimately been served.' Lin's net worth was estimated at 6.8 billion yuan ($941 million).
Yoozoo Games, based in Shanghai, is best known for the Game of Thrones: Winter Is Coming strategy game. Lin Qi acquired the film rights to Remembrance of Earth's Past in 2015. The Netflix adaptation 3 Body Problem premiered in 2024. Xu Yao had led a subsidiary managing the franchise before the fallout.
NASA announced it will hold a competition for the management contract of the Jet Propulsion Laboratory, which the California Institute of Technology has operated since 1958. The current 10-year contract, worth up to $30 billion, expires September 30, 2028. NASA administrator Jared Isaacman said the move ensures accountability and strong taxpayer value as America's space economy evolves. Caltech's president and JPL director said they welcome a fair competition and have been preparing since last summer. JPL, founded by Caltech in 1936, has led historic missions including the Mars Pathfinder rover, the Cassini exploration of Saturn, and the first soft landings on the Moon. The decision mirrors Department of Energy practices of periodically competing management contracts for federally funded research centers.
JPL is a federally funded research and development center managed by Caltech for NASA. It employs thousands of scientists and engineers. The competition could open the door for other universities or private companies to run the lab, though Caltech remains the incumbent with deep institutional ties.
Portland State University plans to cut 52 positions, including a dozen tenured faculty, and eliminate its conflict resolution and university studies departments. President Ann Cudd says enrollment has dropped more than 20 percent since the pandemic, forcing a $35 million budget reduction over two years. The university's infrastructure was built for 30,000 students but now serves 20,000. Faculty unions have issued no-confidence votes, but Cudd says the plan is not final and will incorporate feedback until June 13. Oregon's public universities face widespread financial strain: Southern Oregon University declared financial exigency, and the University of Oregon cut 117 positions. The state's higher education funding is among the lowest in the nation, and the Higher Education Coordinating Commission warns that fund balances could be exhausted within three to five years without changes.
Portland State is an urban public university in Oregon. The state's disinvestment in higher education has left institutions reliant on tuition and reserves. The Oregon Promise Grant, which covers community college tuition for recent high school graduates, may also be ended as lawmakers grapple with funding shortfalls.
Scientists aboard the research vessel Meteor discovered smooth, flat-topped mounds on the Reykjanes Ridge off Iceland that resemble explosive volcanic formations, challenging the long-held view that mid-ocean ridges only produce quiet lava flows. The team, led by marine geophysicist Jonas Preine, used seismic imaging during a test run in June 2024 and found debris deposits suggesting eruptions above the sea surface. The findings could explain historical accounts of islands that appeared and then vanished. Iceland sits on the Mid-Atlantic Ridge, where tectonic plates diverge, but explosive activity was thought impossible in deep water because of high pressure. The discovery suggests that some undersea volcanoes can erupt violently, complicating models of seafloor spreading and volcanic hazards.
Mid-ocean ridges are the Earth's largest volcanic system, typically characterized by effusive eruptions that form new oceanic crust. The Reykjanes Ridge is a segment southwest of Iceland. Expedition M201 was chartered to explore an unexplored deep-water basin dotted with volcanic shapes.
Despite widespread fears that AI is decimating white-collar work, labor data shows no large-scale disruption yet. Unemployment rates for AI-exposed occupations are lower than for less exposed ones, and there is no mass shift to manual jobs. Only one in five companies uses AI in any business function. Erika McEntarfer, former head of the Bureau of Labor Statistics, says AI's impact is small and history shows innovations take time to transform labor markets. The current tough job market for young graduates—with unemployment around 5.6 percent—is driven more by slow hiring post-pandemic than by AI. The piece argues that while AI may eventually disrupt, the data gives time to plan rather than panic.
Recent tech layoffs at Coinbase, Meta, and Cisco have fueled anxiety about AI replacing workers. McEntarfer was fired by the Trump administration after a jobs report that displeased the White House. The Bureau of Labor Statistics continues to show sluggish but stable employment trends.
Uber president Andrew Macdonald said the company is not seeing a clear link between rising AI token consumption and more useful consumer features. Uber spent $3.4 billion on R&D in 2025, up 9 percent, and exhausted its annual AI budget in four months. CEO Dara Khosrowshahi earlier said AI investments are offset by hiring fewer humans. Macdonald noted that if you cannot draw a direct line from AI spending to shipped features, the trade-off becomes harder to justify. The comments reflect growing skepticism in the tech industry about the return on massive AI investments, even as token usage metrics trend sharply upward.
Uber has been integrating AI tools like Claude Code for internal software development. The company's R&D spending also covers autonomous vehicles and delivery logistics. The remarks come amid broader industry debates about whether generative AI is delivering productivity gains commensurate with its cost.
This essay examines the widely reported rise in early-onset colorectal cancer. It notes that while rates in young adults have increased relative to past generations, the absolute risk remains low, and rates in older people have dropped sharply. The author critiques common explanations—ultra-processed food, microbiome changes, environmental exposures—finding most evidence weak or speculative. A key insight: the apparent rise may partly reflect a cohort effect where each birth cohort carries higher lifetime risk, but older age groups still face much higher absolute rates. The piece uses data from Siegel et al. and Downham et al. to argue that the narrative of a mysterious epidemic is oversimplified and that the absolute decrease in older adults is far larger than the relative increase in the young.
Colorectal cancer incidence in adults under 50 has been increasing in many high-income countries, prompting screening guideline changes. The essay is a data-driven look at the statistics behind the headlines, questioning whether the trend is as alarming as portrayed.
A federal judge in Connecticut ruled that AI prompts used by an expert witness to cull documents are discoverable as part of the expert's methodology. In a case against Shell Oil, the plaintiff's expert Dr. Naomi Oreskes used AI to narrow a large document set. The defendants sought the prompts, and the court ordered production, rejecting arguments that prompts were protected as notes or outside discovery scope. The ruling warns that experts using AI should preserve prompts and treat them as part of their methodology, or risk sanctions if representations about their use are later found untrue.
The case is Conservation Law Foundation v. Shell Oil Co. The decision underscores that AI tools in litigation are not a black box; opposing parties can probe how they were used. The ruling may influence how experts document their AI-assisted work in future cases.
