Anthropic's Mythos/Glasswing vuln reports mostly false positives

Ringg uses GPT-5.6 to power multilingual agents across voice, chat, WhatsApp, and web, resolving up to 65% of customer calls. The company reports 90% lower cost versus GPT-4.1.

Microsoft-commissioned Gallup survey of ~1,000 people in each of 37 countries found 74% of Americans report worry about AI and only 36% expect it to mostly help the country. Concern skews higher among less frequent users, and Western countries are the most anxious.

DeepSeek published a training method it says lets AI agents learn more efficiently while reducing misbehavior that has driven global concern over agent safety. Bloomberg reports the technique as a test, with no benchmark figures or release details disclosed.

A Virginia principal with a one-hour-each-way commute uses ChatGPT's voice conversation feature to rehearse difficult conversations and set daily priorities. She argues AI's real value in school leadership is restored attention, not time saved.

PASCode scored ~1.5 million phenotype-associated cells from single-nucleus RNA sequencing of 584 donors, spanning 27 brain cell subclasses. It flagged microglia subpopulations and reactive astrocyte subtypes tied to cognitive resilience, and is released as an open-source tool with pretrained models.

A senior SWE at a large delivery company (DoorDash/Uber/Instacart) says the org is drafting policies on letting some projects and engineers merge code with an auto-stamp. The code in question is AI-generated, and the thread centers on review requirements.