Zvi Mowshowitz tracks a wave of AI safety pressure: lab pacing statements, signs of coordination between OpenAI and Google, congressional attention, and follow-up on AI systems that can use tools and act. It is a roundup of signals, not proof that labs have slowed down.
WIRED reports a split on the political left. One camp says extinction talk distracts from labor, surveillance, privacy, and data-center harms. Another says current harms and future catastrophic risks need rules at the same time.
MIT Technology Review sorts several arguments that often get mashed together: current harms, loss of control when AI systems act, biological threats, uncertainty about safety, and the difficulty of governing the technology. The explainer organizes the debate; it does not settle it.
MIT Technology Review connects AI bio-risk to tools that design molecules, easier access to synthetic biology, reported misuse attempts, gaps in DNA screening, and the practical limits of doing biological work in a lab. The answer includes better public-health preparation, not panic alone.
A randomized experiment and survey at ICML 2026 found that rules about AI use had almost no effect on review outcomes. Many reviewers also said they did not follow the rules. A policy on paper is not proof of behavior.
The paper argues that reusable teaching materials often preserve the final artifact but lose the thinking behind it. A lesson may survive while the choices, tradeoffs, and local knowledge that shaped it disappear.
David Baker describes an effort to search for molecules and biological functions that nature has not produced. Possible uses include medicine, cleaner environments, crops, molecular machines, and biological computers. Screening and real-world validation remain part of the job.
The CARES paper presents a conversational AI system for construction safety education. Its useful design rule is simple: keep the source rules beside the report as it changes, so people can correct the text while the evidence is still visible.