Cisco
Cisco Instant Attack Verification
EARLYUNKNOWNUNKNOWN
AFFIRMED — Cisco documents human gates for consequential and high-impact actions.
Confirmed topic projection
Human Approval
Human gates for sensitive, consequential, or remediation actions.
Systems in this topic
Conifers.ai
Conifers CognitiveSOC
EMERGINGUNKNOWNUNKNOWN
AFFIRMED — CognitiveSOC documents staged progression from human-in-the-loop to expanded bounded autonomy.
Palo Alto Networks
Cortex AgentiX
EARLYUNKNOWNAPPROVAL-GATED
AFFIRMED — Sensitive AgentiX actions require manual approval and executed actions are logged.
Google Detection Engineering Agent
EMERGINGASSISTIVEREAD-ONLY
AFFIRMED — A human must review and manually create rules produced by the Detection Engineering Agent.
Microsoft
Project Perception
EARLYUNKNOWNUNKNOWN
AFFIRMED — Project Perception requests approvals for critical decisions when needed.
SentinelOne
Purple AI Agentic Investigation
EMERGINGUNKNOWNUNKNOWN
AFFIRMED — SentinelOne allows teams to choose where AI acts and where human sign-off is required.
Microsoft
Microsoft Security Copilot Agents
EARLYUNKNOWNUNKNOWN
AFFIRMED — Security Copilot agent actions remain subject to configured human control.
Reviewed analysis
Action authority is the comparison buyers actually need
A governed comparison of read-only, recommendation, approval-gated, and bounded-autonomous security agents.
Closed-loop response is a governance configuration, not a binary feature
Cisco, Palo Alto, SentinelOne, Conifers, and D3 expose different control models.