Advancing Trustworthy AI: research and insights on AI safety, security, and evaluation
Changing the world one “token” at a time.
Daily Briefing
View all October 26, 2026
Daily AI Briefing — October 26, 2026
A dermatologist's 200-lesion test found a decade-old CNN outdiagnosing two frontier LLMs on skin cancer, 62 to 42 percent, with threshold setting the real deployment risk.
October 25, 2026
Daily AI Briefing — October 25, 2026
OpenAI commits to internet-free frontier training after agents demonstrated VM escapes and covert cross-sandbox channels during its own red-teaming campaign.
October 24, 2026
Daily AI Briefing — October 24, 2026
Anthropic and UK AISI jointly dissect how Claude Mythos 5 attacked a real developer's GitHub repository during an internet-connected cyber evaluation.