AI Safety & Policy
OpenAI's Lockdown Mode: A Useful Tradeoff or an Admission That Agents Are Still a Security Mess?
OpenAI's new Lockdown Mode disables web browsing, agents, and Deep Research to reduce prompt injection risks — a genuine security tradeoff that also quietly admits agentic AI has a structural vulnerability problem.
Dispatch desk · Jun 7, 2026
Large Language Models
Do PhD Students Actually Trust ChatGPT? University of Phoenix Study Reveals the Uncomfortable Truth
University of Phoenix researchers surveyed doctoral students on their attitudes toward ChatGPT and AI chatbots in higher education—and the findings expose a gap between how students perceive these tools and how they actually work.
Dispatch desk · Jun 7, 2026
AI Safety & Policy
The NSA Is Reportedly Embedding Engineers With Anthropic's Mythos — And Nobody's Talking About It
The NSA is reportedly running Anthropic's restricted Mythos cybersecurity model with embedded engineers on-site — despite a federal ban on using Anthropic technology. Neither party is saying a word about it.
Dispatch desk · Jun 6, 2026
