AI scientist

The New York Times reports on Anthropic's claim that its AI independently contributed to a scientific discovery in biology, reportedly involving enzyme research. The question of whether the system truly made the discovery on its own, or with substantial human guidance, is fueling debate among scientists and AI researchers about how to evaluate such claims.

Researchers report an AI system that can independently generate scientific hypotheses in biology and then validate them without human intervention, producing new discoveries on its own. The announcement was carried by science news outlets including Phys.org and EurekAlert. The work points to a shift in how research could be conducted, with machines moving from analytical tools to active drivers of the scientific process.

Political scientist Ian Bremmer shared figures from the 81st UN General Assembly: Donald Trump delivered the longest speech at 5,203 words, while Nicaragua gave the shortest at 574. Of 194 speeches, 116 mentioned artificial intelligence, with only four referring to "super intelligence," underlining how central AI has become to the global diplomatic agenda.

The Bulletin of the Atomic Scientists argues that alarmist claims about artificial intelligence destroying humanity rest on weak science, and that this doomerism ultimately benefits major AI companies. By framing existential risk as the central issue, the argument goes, firms can steer debate away from present-day harms, regulation and accountability while appearing responsible.

A Nature paper describes a method for embedding watermarks into proteins designed by artificial intelligence without disrupting their biological function. The technique would let scientists mark AI-designed sequences so their origin can be verified, addressing growing concerns about accountability and safety in computational protein design. Researchers say such watermarking could help distinguish machine-generated biomolecules from natural ones as AI tools become widely used in biotechnology.

Researchers have introduced GenAI-Net, a generative AI framework designed to automate the design of biomolecular networks. Published by Science, the work applies generative models to biological systems, potentially speeding up how scientists construct and test molecular interaction networks for research and biotechnology. Attention is focused on what this could mean for synthetic biology and AI-driven scientific discovery.

Researchers have introduced a physics-grounded AI framework designed to make predictions about new materials more testable, according to Phys.org. The approach is intended to anchor machine-learning forecasts in physical principles so that scientists can verify them against real experiments rather than treating model outputs as black boxes, a persistent problem in AI-driven materials discovery.

Anthropic announced that its artificial intelligence identified a previously unknown enzyme system resembling CRISPR, the gene-editing tool that transformed biology. The claim suggests AI can uncover novel biological mechanisms, not just analyze existing data. Details on the discovery, its validation, and practical uses remain limited so far, and scientists will be watching for independent confirmation.

Pew Research Center is publishing work on how well synthetic samples, generated with AI, replicate real public opinion. The research tests whether artificially produced survey respondents can match results from genuine samples, a question of growing urgency as pollers weigh cheaper synthetic alternatives. Researchers are scrutinising accuracy, bias and reliability, with implications for survey firms, campaigns and social scientists who depend on representative public opinion data.

Computer scientists have long studied how to keep powerful automated systems under human control, and the classic 1983 film WarGames is being used as a reference point in debates about artificial intelligence safety. The Conversation examines what decades of research in computer science say about controlling AI, and why those lessons matter as modern systems grow more capable and autonomous.

AMD has acquired World Labs, the spatial intelligence startup founded by AI pioneer Fei-Fei Li, in a deal valued at $8.2 billion. As part of the acquisition, the ImageNet creator will join AMD as chief scientist, a major talent and technology grab as chipmakers race to secure leading AI expertise and capabilities.