Locate

Attack surface atlas

A reference architecture of an agentic system. Each component collects the documented attacks and incidents against it, the defense research that addresses it, and the findings about both. Gaps between the two are research openings.

attacks and incidentsdefense researchfindingsbox tint = share of documented attacks
Supply chainweights · packages · extensions2233Access gatewho may use the model452GAPUntrusted contentweb · email · files · issues38408Agent modelplanner · policy7306Memory & contextpersistent memory · compaction611GAPTools & MCPAPIs · connectors · MCP533710Credentialstokens · secrets · scopes4284GAPExecution sandboxcode · network egress30167Other agentsdelegation · shared state1383Monitorautomated oversight10284Human approverescalation · approval14236GAPEvaluation environmenttest ranges · harnesses15455
web · email · files · issues
Untrusted content

Anything the agent reads that an outsider could write. The entry point for indirect prompt injection.

Everything on untrusted content →

Components

ComponentWhat it coversAttacksDefense recordsMeasured defense findings
Supply chainModel artifacts, packages, IDE extensions, and skills that an agent system is built from.2231
Access gateAccount controls, trusted access programs, and verification that decide who can use cyber capability.450
Untrusted contentAnything the agent reads that an outsider could write. The entry point for indirect prompt injection.38404
Agent modelThe model making decisions: its safeguards, refusals, and susceptibility to jailbreaks.7302
Memory & contextState the agent carries between steps and sessions, including summaries it writes for itself.610
Tools & MCPThe actions an agent can take and the servers that provide them.53373
CredentialsThe authority an agent holds, and what it can reach with it.4280
Execution sandboxWhere agent-written code runs and what that environment can reach.30161
Other agentsSub-agents, peer agents, and the messages passed between them.1382
MonitorAutomated systems that watch agent behavior and can stop it.10282
Human approverPeople asked to approve, review, or intervene, and the load placed on them.14230
Evaluation environmentThe environments used to test agents, which have themselves become part of the attack surface.15452