Srikumar and Leibowicz studied four systems that let AI use tools to do tasks. They say key records are missing: who ran the system, what permissions changed, and when a person stepped in. An approval is hard to audit if no one can check what happened next.
Van and Tran test what happens when a source disagrees with what an AI already knows. Their method can choose either answer or hold back. Its test gains partly depend on knowing how often the source is right, a number real users may not have.
An early research abstract describes a game with 42 users across 1,260 rounds. AI predictions based on conversation did not beat those based on past choices overall. The finding is about that game, not whether workers will use a new tool.
Grist reporting carried by WIRED shows how an AI-drafted warning can sound clear while leaving blanks or getting a location wrong. A qualified person still needs to check facts, recipients, and instructions against official sources before any real alert goes out.
Jason Parham tried a health app that predicts how long someone might live. His estimate changed with questionnaire answers, and he describes more anxiety and follow-up prompts. His experience does not show whether the app helps or harms people overall.
WIRED spoke with computer scientists after OpenAI and Anthropic published claimed AI-assisted proofs challenging a long-held theory. Researchers describe excitement and the cost of reading a flood of new work. The proofs still need independent checks and a clear explanation.
Ampomah and Le study user reviews of third-party bots on Discord. Their research abstract points to privacy concerns for members who did not install a bot but may still have data visible to it. Reviews do not prove what any one bot did.