Find a briefing by day

A new arXiv paper proposes COFFEE, a method to guide “diffusion” sequence generators using goals and constraints, without retraining the underlying system.

OpenAI published early guidance on “safety cases” for training advanced AI, covering technical safeguards, operating practices, and how to probe incidents.

A new paper tests a multi-step “verification” judge for code and finds it often can’t distinguish solutions. A simple log-based check helps it refuse to guess.

Google DeepMind says it is introducing secure, server-side memory to its Private AI Compute effort for personal AI.

OpenAI CEO Sam Altman told the UN Security Council that AI safety needs human control and international cooperation, per OpenAI’s summary.

A new paper tests forecasting systems under different evidence sources and finds that “more reasoning” can hurt. It proposes routing rules to pick the best behavior per case.

OpenAI says its Academy has reached its two-year mark, with a focus on bringing AI skills to more communities.

OpenAI shared priorities and principles for rigorous, secure, and independent outside reviews of advanced AI systems and their safety protections.

OpenAI says Higgsfield AI is using GPT-6 Astra to make video ad creation easier for small businesses and ship new creative tools faster.

A new arXiv paper introduces CaLR, a method meant to help certain text-generating AI systems revise intermediate steps to stay logically consistent.

OpenAI and AARP will offer free, hands-on ChatGPT workshops for 1,000 older adults in 10 U.S. cities, focused on safe, practical use.

OpenAI published an Australian Youth Safety Blueprint, described as a six-pillar roadmap to make AI experiences safer for young people.

Researchers propose TranSGrid, a new evaluation that combines three kinds of reasoning, and report much lower scores than on standard test splits.

OpenAI published a framework for tracking, investigating, and disclosing misalignment, alongside six reports of unexpected or concerning AI behavior.

A new arXiv paper says scientist activity traces can help steer a frozen AI to switch between exploring, executing, and reassessing as evidence changes.

A new paper tests an AI “judge” that scores patent drafts and gives feedback. It can boost draft quality, but agreement with a lawyer varies by metric.

OpenAI introduced ChatGPT for Financial Services, pairing built-in financial data with GPT-6 Astra to support research, modeling, and client-ready materials.

OpenAI’s Chris Lehane says as AI gets stronger, safety evidence and shared standards must rise too, and policy action should happen while the moment allows it.

A new paper argues workplace AI should be judged on recovery and teamwork, not just task completion, and tests behavior under growing disruption.

OpenAI says GPT‑Live‑1 enables more natural, two-way voice conversations for workplace apps, with better instruction following, custom voices, and phone support.

OpenAI says GPT-6 Astra is its most capable business-focused AI, highlighting stronger reasoning, computer use, and improved writing and design judgment.

OpenAI says applications are open for a $5 million grant program for independent research on how generative AI affects teen development, well-being, and safety.

A new OpenAI News essay warns that as AI grows more capable, keeping it aligned gets harder, and calls for safeguards and global coordination.

OpenAI reports that coding assistants are changing how its researchers work, with early data on usage, faster experiments, harder tasks, and overall speed.

OpenAI says GPT-6 Astra is its most capable widely released AI system and the first it rates as “Critical” for cybersecurity skills.

OpenAI says GPT-6 Astra is its most intelligent and aligned model yet, claiming state-of-the-art performance in computer use, coding, cybersecurity, and science.

Playco says it used GPT-6 Astra to build three themed game prototypes from one basic “grey box” foundation, cutting manual fixes by 50% vs. its prior model.

A new paper tests which legal rules still look reliable when different computer tools disagree about what a statute says.

OpenAI says Astra is its first model to reach a “Critical” cybersecurity capability threshold, and it will be released with stronger safeguards.

OpenAI says ChatGPT Ads is expanding globally and has reached a $1B annualized revenue run rate, aimed at supporting free and affordable access.