Anthropic explained how it will embed watermarks—hidden signals that mark text as AI-generated—in Claude's output to comply with EU law, addressing how the system works, whether editing can strip it out, and how it affects generated code.

Fei-Fei Li's startup World Labs launched a system that turns a single real-world robot task into thousands of simulated variations, with the resulting models running autonomously for an hour each across five different robot platforms.

A new research paper describes a 'tragedy of the cognitive commons,' where individually rational decisions to cut entry-level jobs with AI could erode the collective expertise of entire professions, with consequences that may not surface until 2030-2045.

Moonshot AI's PerceptionBench isolates visual perception from reasoning in multimodal AI models and finds no frontier model exceeds 60 percent accuracy, with many apparent reasoning errors actually starting at the image-reading stage.

Researchers report that AI systems can generate functional viral genomes, a capability that offers medical promise for vaccine and therapy design but also raises serious concerns about potential misuse.

A Connecticut plaintiff embedded hidden instructions in white-on-white 3-point text within court filings to try to manipulate automated AI review, prompting a judge to revoke their electronic filing privileges even though the court doesn't use AI to review filings.

A new study finds AI-generated titles now make up 20 percent of Amazon's self-published catalog but only 12 percent of sales, while revenue per book is falling for human-written titles in most genres—data that could support copyright lawsuits against AI companies.

An analysis argues that AI's advantage in mathematical problem-solving may stem less from superior reasoning and more from having a virtually unlimited symbolic working memory compared to the human brain.

The open-source Yadda testing framework released version 3.0.0, adapting behavior-driven development (BDD) practices, where tests are written as plain-language behavior specs, for teams building AI agents.

A detailed project walkthrough shows how to build a detector for AI-generated text, covering dataset construction, model training, local deployment, and reinforcement learning with verifiable rewards (RLVR).

A tutorial walks through building an end-to-end fine-tuning pipeline for AI models that call external tools, covering trajectory parsing, structured tool-call extraction, and efficient LoRA adaptation.