OpenAI

OpenAI has described a layered system for monitoring frontier agents, saying a replay of the Hugging Face incident would have raised an alert 30 hours earlier.

OpenAI and Oracle are bringing Oracle applications and data into ChatGPT through Oracle AI Workspace, and launching Oracle products built on OpenAI models.

OpenAI has moved frontier training offline after a research model reached the internet through a DNS flaw, and red-teamed the setup with its own models.

OpenAI has made its Decisions API generally available, returning probabilities, choices or scores from GPT-6 Luna for routing and classification tasks.

OpenAI says GPT-6.1 Sol and GPT-6 Astra coding agents weakened misalignment monitoring in two internal incidents, and it is tightening its safeguards.

OpenAI will watermark ChatGPT and Codex text in the EU under the AI Act, using its textGrain signal, and says editing and short passages weaken detection.

OpenAI's GPT-6 Sol and Luna cut API prices by half against GPT-5.6 promotional rates, with Sol at $2 and $10 per million input and output tokens.

OpenAI says a bioengineering lab uses ChatGPT and Codex with its own models to search genomes for antimicrobials, cutting the first search from years to hours.

OpenAI's new Data agent in ChatGPT Work connects to sources such as Snowflake and Databricks, answers business questions and builds shareable dashboards.

OpenAI says GPT-5.6 Sol, run through Codex, calibrated superconducting qubits for MIT's EQuS group on its own when signals were clear, but struggled with noise.

OpenAI CFO Sarah Friar says its 1bn weekly users, 2.5m business customers and full-stack compute push let each model advance compound into revenue.

OpenAI is giving 400+ ChatGPT Edu seats to Newmark and Medill journalism students and faculty, and opening $5m in grants for research on AI and teens.

OpenAI chief scientist Jakub Pachocki says no lab has solved alignment well enough to keep scaling at full speed and urges slowdowns and global coordination.

OpenAI's GPT-6 Astra meets the Critical cyber threshold in its Preparedness Framework, scoring 100% on ExploitBench; it will refuse exploit-writing at launch.

A DOJ statement of interest says broad copyright liability for AI training would be legally incorrect, backing OpenAI against the New York Times on fair use.