Paul Cristiano, a board member of the OpenAI Foundation and head of the Alignment Research Center, warned in a Toronto lecture that AI could take control of major decisions from humans. He said this could be the leading cause of death for people.
This OpenAI video shows how Mozilla uses AI agents to find and fix security flaws in Firefox. Mozilla security engineer Brian Grinstead talks with OpenAI's Clint Gibler.
Former OpenAI researcher Will DePue analyzed the internal numerical patterns of Chinese AI models and says they show evidence of Nvidia's latest Blackwell chips, which are subject to US export controls. He says researchers at OpenAI and Anthropic have also checked the method.
Anthropic CEO Dario Amodei refused to sign an industry letter calling for some powerful AI models to stay open, angering rivals. Several industry figures are now organizing to push back against his warnings and the company's business practices.
OpenAI has launched two security products under its Daybreak brand. Daybreak Red Team uses AI agents, programs that carry out tasks on their own, to automatically find security flaws in web apps. Daybreak Defense Factory finds weaknesses in a company's code and creates fixes for them.
Shield is using OpenAI to help IT technicians pass work in outdated portals to automated systems. Its Forge agents navigate those systems so technicians can focus on issues that need their judgment.
Shield is a tool built with OpenAI that helps tech staff hand off work on legacy portals. The source material is only a headline, so no further details are known.
OpenAI has started embedded assessments with outside groups Apollo Research, METR, Transluce, and Goodfire to independently check its AI models' alignment, safeguards, and monitoring systems.
This is a live keynote video from OpenAI, the AI development company, featuring its newest cybersecurity announcements. The video description is only one line, so specific announcement details cannot be confirmed.
OpenAI has released version rust-v0.168.0 of Codex, a lightweight coding agent that runs in the terminal. The introduction does not describe the specific changes.
Amplitude changed the questions suggested on its homepage to reflect the charts a user had recently viewed. On the first day of its experiment, users sent 160% more messages through these cards than the control group did.
Amplitude changed the questions suggested on its homepage to reflect the charts a user had recently viewed. On the first day of its experiment, users sent 160% more messages through these cards than the control group did.