Browse by guarantee
Every entry is tagged against the white paper's framework, not a generic "agent safety" scheme. The top-level axis is SPRS: the property a system has to hold to be trusted.
Security
Resisting adversaries: prompt injection, tool abuse, supply-chain and execution-shell attacks.
Browse →Privacy
Protecting data the agent reads, remembers, and emits across context and memory.
Browse →Reliability
Doing the right thing under uncertainty: planning, coordination, evaluation, and recovery.
Browse →Safety
Bounding consequence: oversight, permissions, intent alignment, and transactional agency.
Browse →Two tracks, one taxonomy, one review gate
The resource holds two asset classes with opposite physics. Both write the same schema, carry the same taxonomy tags, and pass the same human review before publishing.
Resources
Classics, courses, open source, and peer-reviewed research, human-curated and filterable by harness layer and SPRS guarantee.
Browse the map →Incidents
A primary-source-verified log of what actually goes wrong in deployment, classified by the harness layer the failure exposed. Open to community reports.
See incidents →