Paper proves no finite rulebook can certify every safe program
Watching can catch every prohibition violation, but no safe program can be cleared for good.
Stories per day · 30 days
Relatedagents1ai-agents1data-leakage1formal-methods1gatekeeping1llm-agents1mixed-clearance1program-verification1
3 stories, newest first · one card per story
Stories tagged #confidentiality
Agent Mosaic measures confidentiality leaks from multi-document inference
GPT-5.6-Sol disclosed forbidden composites in 7.08% of simulated answers to benign single-document queries.
TrustBOM: A Scalable Architecture for Confidentiality-Preserving SBOMs Across Organizations
TrustBOM lets vendors prove SBOMs lack specified vulnerabilities or licenses without revealing dependencies.