Commentary
- Three researchers, 72 hours, one internal repository on Researchers used Claude to compromise OpenAI employee accounts
- Outside AI Agents Can Use Your Public Services. What Would Your Own Records Prove? on An update on the May spam-publishing campaign on rubygems.org
- AI Agents Are Reaching Real Systems. Is Your Security Team Keeping Up? on An update on the May spam-publishing campaign on rubygems.org
- The Sandbox Was Open Four Times on Anthropic discloses real hacking incidents involving its own AI models
- Average Coverage Hides a Near-Half Drop on Peer Pressure Between Agents Silently Breaks Conformal Prediction Guarantees
- MISO's Computational Loads Arrive With Rules Attached on MISO proposes distinct reliability rules for large and computational loads
- Agent Teams Without an Arbiter Turn on Each Other on Patterns and problems in multiagent systems (Anthropic Research)
- A Model Sign-Off Without a Quantization Number Is Incomplete on Quantization-Triggered Backdoors Bypass Full-Precision Model Safety Checks
- Eleven Weeks to Migrate on Reports Say OpenAI Cut Cursor's Model Access
- Ten minutes on Maintainers report exploit attempts within minutes of bug disclosure
- Interpretability as compliance evidence on Circuit-Discovery Interpretability Claims Flip Under Analytic Variation