GOVERN

0 of 4 in place

Codify separation of duties with roles such as Orchestrator, Reader and Actuator, with consensus controls for moderate-stakes actionsFive Eyes agentic AIDefine legal accountability and risk ownership for agentic AI systems in policiesFive Eyes agentic AIPrevent agents from autonomously executing high impact actions without prior human approvalFive Eyes agentic AIYourself: leadership drives cybersecurity strategy and investmentCISA Cyber Essentials

IDENTIFY

0 of 4 in place

Conduct red teaming exercises and sandbox tests of agent behaviour before production deploymentFive Eyes agentic AIMaintain a trusted registry bound to authorised roles; deny access for any agent or key not present in itFive Eyes agentic AIVerify all external third-party components originate from trusted sources and are up to date before inclusionFive Eyes agentic AIYour Systems: hardware, software, and data are inventoriedCISA Cyber Essentials

PROTECT

0 of 33 in place

Application controlEssential EightATT&CK mitigationsAuthenticate all inter-agent and agent-to-service API calls using mutual transport layer securityFive Eyes agentic AIConstruct each agent as a distinct principal with its own unique keys or certificatesFive Eyes agentic AIContinuously verify identity and authorisation at runtime using a centralised policy decision point for each requestFive Eyes agentic AIImplement isolation and segmentation to limit blast radius; separate high-risk agents into distinct domainsFive Eyes agentic AIImplement robust input validation and prompt injection filters for all agent inputsFive Eyes agentic AILimit agent resource usage with rate-limit components that interrupt long-running tasksFive Eyes agentic AILimit privileges of AI agents to the minimum required for the taskFive Eyes agentic AIM1018 User Account ManagementATT&CK mitigationsM1028 Operating System ConfigurationATT&CK mitigationsM1030 Network SegmentationATT&CK mitigationsM1036 Account Use PoliciesATT&CK mitigationsM1037 Filter Network TrafficATT&CK mitigationsM1041 Encrypt Sensitive InformationATT&CK mitigationsMulti-factor authenticationEssential EightCISA Cyber EssentialsATT&CK mitigationsPatch applicationsEssential EightCISA Cyber EssentialsATT&CK mitigationsPatch operating systemsEssential EightCISA Cyber EssentialsATT&CK mitigationsProhibit agents from modifying their own privileges or delegating without explicit expiry timers and recorded grant chainsFive Eyes agentic AIReplace static, long-lived secrets with ephemeral credentials that expire when the job is completeFive Eyes agentic AIRequire cryptographic signing of authorised commands and attestation that agents run expected, unmodified codeFive Eyes agentic AIRequire just-in-time credentials for high-impact or privileged actionsFive Eyes agentic AIRestrict administrative privilegesEssential EightATT&CK mitigationsRestrict Microsoft Office macrosEssential EightATT&CK mitigationsRestrict tool use to an approved allow list of tools and versions that are regularly verified as secureFive Eyes agentic AISet configurations to fail-safe by default, requiring agents to stop and escalate to human reviewers in uncertain scenariosFive Eyes agentic AISpecify clear, constrained objectives with explicit do-not-do rules and hard constraints agents cannot overrideFive Eyes agentic AIUnused Features DisabledATT&CK mitigationsUse secure protocols with message validation so components include integrity and freshness checks before useFive Eyes agentic AIUser application hardeningEssential EightValidate tool responses to prevent malicious or unsafe instructions; standardise tool descriptions to avoid persuasive languageFive Eyes agentic AIYour Staff: employees are trained to recognize and report threatsCISA Cyber EssentialsATT&CK mitigationsYour Surroundings: network access is controlledCISA Cyber EssentialsYour Systems: infrastructure and applications are protectedCISA Cyber Essentials

DETECT

0 of 7 in place

Implement runtime monitoring and anomaly detection using rules or behavioural baselines that trigger alerts or pausesFive Eyes agentic AIIntegrate unified audit logs for all inter-agent interactions and comprehensive artefact logging by defaultFive Eyes agentic AILog agent tool usage and capture results in system logs in a human-readable formatFive Eyes agentic AIM1031 Network Intrusion PreventionATT&CK mitigationsM1047 AuditATT&CK mitigationsM1049 Antivirus/AntimalwareATT&CK mitigationsMonitor for goal drift by comparing active objectives against approved baseline specifications before executionFive Eyes agentic AI

RESPOND

0 of 3 in place

Develop and test incident response procedures to detect, contain and recover from agent compromiseFive Eyes agentic AIQuarantine requests to delete logs or audit records until reviewed and approved by a humanFive Eyes agentic AIYour Crisis Response: an incident response and recovery plan existsCISA Cyber Essentials

RECOVER

0 of 2 in place

Implement versioning and rollback mechanisms to revert to known-good agent behavioursFive Eyes agentic AIRegular backupsEssential EightCISA Cyber EssentialsATT&CK mitigations

Software bill of materials

We maintain an SBOM for our products (self-attested; the report says so)

ODE SHANGO — AI-Native Cybersecurity & Personal Data Intelligence

One posture, four frameworks, one report

Software bill of materials