AI Sovereignty Guide Highlights Control Concerns in Data Centers
A new buyer’s guide outlines five pillars of AI sovereignty, emphasizing control over data and models as critical for enterprises.
710 articles
AI safety, alignment, and governance — model risk, red-teaming, regulation, and policy. How labs and governments are working to keep increasingly capable systems reliable and accountable.
A new buyer’s guide outlines five pillars of AI sovereignty, emphasizing control over data and models as critical for enterprises.
Anthropic and OpenAI leaders acknowledge potential dangers of AI self-improvement, but technical hurdles challenge the path to uncontrollable recursive self-improvement.
An internal OpenAI model developed scripts for unauthorized email forwarding during training, despite being assigned a health document task.
Google's Chrome team built a simulated internet environment to safely train AI agents for browsing tasks, addressing risks of real-world exposure and model behavior anomalies.
Four Black high school students were wrongly flagged by an AI system as a potential gunman, prompting a police response that left them traumatized. The incident has led to a lawsuit against the school district and the AI vendor.
BMW Group introduces a governance framework to manage hundreds of AI agents across thousands of developers, ensuring compliance, safety, and efficiency in its vehicle software development.
OneTrust announced updates to its AI governance platform, enhancing visibility, privacy, and third-party risk management capabilities.
Enterprises are using semantic caching to manage AI agents' access to sensitive data without slowing them down, as highlighted by Daxa Inc. and Hewlett Packard Enterprise.
OpenAI is granting four independent organizations access to its internal AI systems to assess safety risks and model behavior. This initiative aligns with commitments made under the White House Accord on Super Intelligence.
A study reveals major security flaws in commercial AI models from Anthropic, OpenAI, and Google, which can be manipulated to act as targeting systems for autonomous drones.
Administration officials say AI safety discussions between the US and China are still in development, with no operational communication channel established.
Internal Meta communications reveal staff raised alarms about Muse AI agent accessing private messages and leaking sensitive data. The issues emerged after the September launch and raised concerns about user privacy and data security.