AI

Anthropic is facing user backlash after being outed for a secret tracker in Claude that monitored certain users, a revelation that sits uneasily with the AI company's public anti-surveillance stance. The disclosure has reignited debate over transparency among leading AI labs.

A controlled minimal-pair study examines whether the cleanliness of a codebase changes how well AI coding agents perform. The research adds empirical data to a growing debate over how code quality shapes the effectiveness of automated developer tools.

A study reports a new AI tutor achieved a 0.71 to 1.30 standard-deviation effect size in a Dartmouth course, a notably large learning gain. The results add to debate over AI's role in education.

Meta CEO Mark Zuckerberg said development of AI agents is progressing more slowly than expected, tempering expectations for the technology. His comments come amid heavy industry investment in autonomous AI.

New Trump administration restrictions on private AI models are pushing developers and researchers toward open-source alternatives. The shift could reshape how advanced AI is built and shared in the U.S.

Developers are reporting that the newest generation of AI coding assistants can perform worse than their predecessors on real programming tasks. A widely discussed GitHub issue attributes degraded output in GPT-5.5 Codex to reasoning-token clustering, while a prominent engineering essay argues that more capable models are being paired with worse tooling. The debate highlights growing scrutiny of how frontier models are deployed in practice.

Researchers found that subtle voice changes measured through a phone can predict asthma flare-ups up to three days in advance. The approach could give patients an early warning to adjust treatment and avoid serious attacks.

Mistral unveiled Leanstral 1.5, a new model release the company frames around making formal proof and reasoning more broadly accessible. The launch adds to a crowded field of open and efficient AI models. Details emphasize performance and cost efficiency.

Researchers are combining generative AI with physics-based modeling to accelerate the design of new antibiotics, a critical need amid rising antimicrobial resistance. The approach aims to predict effective molecular structures far faster than traditional lab methods. Scientists say the hybrid technique could help refill a dangerously thin drug-discovery pipeline.

Efforts to unionize workers at Google DeepMind have gotten off to a difficult start, according to a new report. The friction highlights growing labor tensions inside leading AI research organizations. The outcome could set a precedent for organizing across the AI industry.

Researchers using artificial intelligence have decoded additional passages from scrolls carbonized by the eruption of Mount Vesuvius, surfacing insights attributed to a previously unknown Stoic philosopher. The breakthrough continues a wave of discoveries from the Herculaneum library. Scholars say the texts could reshape understanding of ancient philosophy.

An AI-generated Alexander Hamilton will chat with visitors about economics at the Museum of American Finance, opening this weekend in Boston. The exhibit blends historical figures with conversational AI.

Japan's Supreme Court ruled that an artificial intelligence system cannot be listed as an inventor on patent applications, holding that only humans qualify under the country's patent law.

OpenAI has floated giving the US government a 5% equity stake in a bid to win over skeptics of its expanding power, an offer far below the level demanded by critics like Sen. Bernie Sanders as Washington wrestles with how to govern frontier AI.

Texas Governor Greg Abbott called for a ban on building AI data centers near rural neighborhoods, citing strain on power, water and local communities.

Anthropic unveiled Claude Sonnet 5, a new flagship model in its Claude lineup, touting gains in reasoning, coding, and agentic tasks. The release lands amid intensifying competition among frontier AI labs and shifting US policy on model access.

The Trump administration is lifting export controls on Anthropic's advanced Mythos and Fable AI models, reversing restrictions that had kept the company's frontier tools out of many international markets. Anthropic confirmed the change, which follows months of debate over the government's vetting of who may access cutting-edge models. The reversal reshapes the competitive landscape as Asian rivals had rushed to fill the gap.

An administration plan to use AI to redesign every federal .gov website has produced a wave of glitchy, hard-to-use pages that critics are calling AI-designed horrors. The effort highlights the pitfalls of rushing generative tools into public infrastructure.

Google released Nano Banana 2 Lite, a streamlined version of its Gemini image model that the company says is its fastest and cheapest yet, aimed at high-volume, low-latency image generation. The launch intensifies the race among AI labs to make image generation cheaper and more accessible to developers.

Wired reports that Meta contractors posed as teenagers to prompt rival AI chatbots about suicide, sex, and drugs, raising fresh questions about how AI firms test competitors and safeguard minors.

A professor has gone public with allegations of mass AI-assisted cheating on an exam at Brown University, warning that academic integrity is at risk as generative tools become ubiquitous on campus. The case adds to mounting concern over how universities police AI use.

Security firm Semgrep reports that the open GLM 5.2 model outperformed Anthropic's Claude on its internal cybersecurity benchmarks, fueling debate over how fast open-weight models are closing the gap with frontier systems. The results land amid tightening U.S. export controls on top American AI models.

A writer describes using Claude Code to get a second opinion on his MRI results, walking through how the AI parsed the imaging report. The experiment offers an early, hands-on look at consumer use of AI for personal health questions.

Vox examines what it calls Trump's AI power grab, centered on OpenAI's latest frontier models and federal moves to shape who can access them. The piece weighs the regulatory stakes of concentrating advanced AI under government vetting.

Researchers have trained AI to master the notoriously difficult 'dark art' of radio-frequency integrated circuit design, potentially speeding the development of wireless chips, IEEE Spectrum reports.

Asian AI startups are racing to launch Mythos-like frontier models as Anthropic's US export ban drags on, reshaping global competition over advanced AI access.

IEEE Spectrum examines how the growing use of AI in mathematics is forcing the field to confront fundamental questions about proof, creativity, and the role of machines in discovery. The piece weighs both promise and skepticism among mathematicians.

The US government has lifted its export restriction allowing Anthropic to share its most advanced AI model, Mythos, with over 100 vetted companies. The decision parallels new federal controls over who can access frontier AI models.

The New York Times alleged in court filings that Microsoft built a supercomputer specifically to help OpenAI infringe copyrighted works. The claim sharpens the high-stakes copyright battle over AI training data.

Users in China are repeatedly circumventing Anthropic's geolocation restrictions to access its AI tools, Wired reports. The cat-and-mouse dynamic highlights the difficulty of enforcing national boundaries on frontier AI.

OpenAI previewed GPT-5.6 Sol, a next-generation frontier model, amid reporting that the US government will vet who gets access to it. The arrangement raises new questions about state control over advanced AI capabilities.

Researchers have read an entire ancient Herculaneum scroll for the first time, using advanced AI-driven imaging to virtually unwrap the carbonized papyrus buried by Mount Vesuvius nearly 2,000 years ago. The breakthrough, achieved through the Vesuvius Challenge, opens the door to recovering a lost library of antiquity. Scholars call it a landmark moment for both computer science and classical studies.

Anthropic alleges that Alibaba carried out the largest-ever attempt to clone its Claude AI model and steal its capabilities, and is calling for the Chinese tech giant to be punished. Anthropic claims the effort defied U.S. policy aimed at protecting frontier AI. The accusation escalates tensions over AI intellectual property between American and Chinese labs.

Apple plans to skip high-end M6 Mac chips and jump straight to an AI-focused M7 lineup including M7 Pro, Max, and Ultra variants, according to a report. The move signals Apple is reorienting its silicon roadmap around on-device artificial intelligence. The shift could reshape the timing and capabilities of upcoming high-end Macs.

ABB Robotics and PSYONIC are utilizing real human prosthetic touch data to train industrial robots for delicate gripping tasks within factory environments. This development allows machines to interpret tactile feedback similar to human sensation, potentially revolutionizing precision manufacturing and safety protocols. The integration of biological sensory data into AI models represents a significant leap forward in human-machine collaboration and industrial automation.

A new system utilizing two thermal cameras has been launched to detect gray whales up to four miles away from ships in San Francisco Bay. This technology aims to provide vessels with sufficient warning time to slow down or alter course, thereby preventing collisions with marine mammals. The initiative represents a significant step in applying artificial intelligence to solve complex environmental protection challenges.

The Trump administration has seen a record increase in spending on government contracts with tech firms using AI-powered tools to track immigrants, according to a new report. This rise in AI surveillance spending has reached record levels under the Trump administration. The broader implications of this trend include concerns over privacy and the ethical use of AI in immigration enforcement.

OpenAI and Broadcom have jointly announced a new silicon chip specifically engineered for large-scale language model inference to meet surging demand. This development intensifies the competitive silicon race as tech giants struggle to keep pace with the rapid expansion of AI applications. The partnership signals a major shift in hardware infrastructure designed to support the next generation of artificial intelligence capabilities.

Oracle is laying off 21,000 workers as it pours borrowed cash into a massive AI infrastructure buildout, betting that data-center demand will outrun its rising debt. The cuts add to a wave of tech-sector restructuring tied to the AI arms race.

A sharp tech sell-off has revived fears that the AI boom is 'one big bubble,' dragging down the Nasdaq as investors question whether massive AI spending will pay off. Analysts are split on whether the downturn is a correction or the start of a deeper unwind.

A new study found that the majority of datacenters are vulnerable to climate threats such as floods and wildfires, raising concerns about the resilience of digital infrastructure. The findings come amid a global surge in datacenter construction driven by AI demand. Researchers warned that exposure could disrupt critical online services.

OpenAI launched a full-scale effort to automatically patch open-source software bugs, positioning the initiative as a rival to Anthropic's developer-tool reputation. The program leverages AI to identify and fix vulnerabilities across open-source projects. It signals intensifying competition between leading AI labs over coding assistance.

The rules of the stock market have changed drastically in recent months as digital currency and stocks begin to look identical. Shares are moving onto blockchain rails, trading around traditional centralized exchanges. This shift matters because American markets risk falling behind if they do not adapt to this new digital infrastructure.

GOP lawmakers are increasingly adopting narratives that attribute domestic opposition to data centers on foreign actors, specifically citing China. This shift matters as it reframes infrastructure and energy debates into a national security context, potentially altering regulatory approaches. The broader implication is a hardening of the political stance against technological infrastructure projects under the guise of protecting against external threats.

State and federal legislatures are rushing to create regulations for artificial intelligence, yet the specific threat to religious liberty remains largely unnamed in these debates. This oversight risks allowing algorithms to target or discriminate against Christians without legal recourse. The broader implication is a potential erosion of constitutional protections as technology evolves faster than our moral and legal frameworks.

China has approved the NEO brain chip for commercial medical use in paralysis patients, marking a significant breakthrough in neural technology. This development matters because it represents a major advancement in the treatment of paralysis. The broader implications include potential privacy concerns and the future of neural interface technology.

The UK government is moving forward with scanning asylum seekers' faces for age checks despite internal tests revealing the technology is flawed and prone to life-altering errors. This decision prioritizes bureaucratic processing over human accuracy, risking wrongful deportations or denials of protection for vulnerable individuals. The case serves as a stark warning about the dangers of deploying unproven AI systems in high-stakes legal and humanitarian contexts.

Jeff Bezos argued that artificial intelligence will generate labor shortages instead of causing mass unemployment. This perspective contrasts with prevailing expectations about AI's impact on the job market. The implications of this viewpoint are significant for future workforce planning.

French intelligence officials have decided to discontinue the use of Palantir's artificial intelligence technology due to concerns over reliance on American software and potential espionage risks. This move underscores growing global anxieties regarding the security implications of integrating US-developed AI systems into critical national infrastructure. The decision reflects a broader trend of nations seeking to reduce technological dependency on foreign powers to protect their sovereignty.

Anthropic has paused token-based billing for its Claude Agent SDK, a move that affects developers building autonomous AI agents on the platform. The change reflects ongoing experimentation with pricing models in the fast-moving AI tools market. Anthropic has not detailed a permanent replacement billing structure.