Decision Trace records the task, evidence, agent contribution and human decision
22%
Of MITRE ATT&CK techniques enterprise SIEMs have detections for (CardinalOps 2025)
33.4B
Events analysed monthly in a Fortune 500 environment: the evidence scale one ledger must cover
92.5%
Incidents resolved by AI under human governance, with <20 min mean time to detect and respond
Copilots suggest. Playbooks script. Agents act, and someone has to govern that. The 92.5% figure is that control plane running our own agents; the same plane governs everyone else's. Denominators: 92.5% = cases resolved without human escalation ÷ cases opened, rolling 30 days, all customer environments; <20 min = aggregate across managed-service environments; 22% per CardinalOps 2025 State of SIEM Report.
Trusted by Fortune 1000, FTSE 100, public-sector, financial services, and healthcare teams.
WATCH
AI governance starts with knowing what is running.
A short film on why AI governance starts with knowing which AI tools run in your business and what data they can reach.
SenseOn brings AI under control · 2 min 14 sec · Captions availableRead transcript
Right now, every team in your business is using AI. The problem? Almost no one can tell you which tools, with what data, are making what decisions.
The first wave was about adoption, and it worked. 88% of organisations now use AI. Generative AI reached 53% of people in three years, faster than the PC or the internet.
But every new tool is a new blind spot. 97% of firms breached through their AI lacked proper AI access controls. 63% have no AI governance policy, or are still writing one. Reported incidents jumped 55% in a year.
Regulation is catching up. The EU AI Act sets hard obligations for high-risk AI, with most applying from December 2027.
So the question: do you actually know how many AI tools are running inside your organisation right now, and which ones are a real risk?
You do not have to manage this blind. SenseOn gives you a clear view of the AI tools your business uses, and the governance to keep it safe.
Ask Horus and it answers instantly. Which AI tools run across the business? Seven, all named. Which hold OAuth access to your Microsoft 365 data? Three, flagged. Which ones need a closer look? Two, flagged on the register. AI governance you can finally see.
While Horus answers what you ask, Insight reports weekly on the AI tools in use across your business. Shadow tools, ungoverned code, quiet access to your data. Insight brings it all into view, turning scattered AI risk into something you can govern.
That is how SenseOn brings AI under control. Ask Horus anything. Let Insight report on the rest, every week.
WHY NOW
Approval is not control when software can act.
Your board approved AI adoption; your team still needs to prove what consequential workflows did with identities, data, code and business systems.
Adopting safely means holding one reconstructable chain per action: context, authority, outcome and correction.
SenseOn connects provider, identity, tool, network and target-system evidence so a team can find missing authority, contain access and prove whether the result was reversed.
Security for AI governs consequential AI workflows across the estate. AI for Security applies governed AI to defined security operations work. Both rely on source-linked evidence, explicit authority and human decisions.
Observe and control the AI providers, agents, identities, tools, data access, and code changes already running across your estate. You are on this page.
FIVE CISO PRIORITIES
Five decisions now need evidence.
Agent risk is an identity, supply chain, data, and action problem. Each priority is a question your evidence has to answer before the board signs off, and again when something goes wrong.
Which agents, providers, and tools are approved, and who used what?
Provider consoles answer one workspace at a time. The SenseOn Data Fabric assembles the estate-level answer from identity, endpoint, network, cloud, SaaS, and provider evidence on your retention terms. Decision enabled: is this AI use approved, and under which policy?
What sits between a prompt and a production change?
Models, packages, plugins, and knowledge sources each log only their own half of the hop. The Intelligence Fabric joins them into one source-linked chain across repo, CI/CD, runtime, and identity evidence. Decision enabled: which hop introduced this change?
Which identities, tokens, and tool scopes can an agent exercise?
An agent's blast radius is its scope grant, not its model. The Agent Control Plane holds identity, tool scopes, policy, approval, timeout, and escalation for every agent that acts. Decision enabled: who can approve escalation, and what should be revoked?
What sensitive data moved through an AI workflow?
Prompt-level DLP sees the input; the estate sees where the output went. SenseOn correlates model context with endpoint, network, SaaS, and code evidence on one chain. Decision enabled: did protected data leave a controlled system, and what must be contained?
Can responders reconstruct and reverse what an agent did?
Decision Trace records the task, evidence, agent contribution and human decision. Containment and rollback stay with named owners. Decision enabled: can you undo what the workflow did, and prove it?
Provider logs and posture tools record their own world. Nobody reconstructs yours.
Provider audit logs account for their own workspace. SenseOn connects what each source contributes with the identity, network, tool and target-system evidence around it. Your Bedrock, Microsoft, GitHub and AI-posture investments stay in place.
Provider
The provider-log era: native logs and AI posture tools
Connected action evidence
AI security posture and prompt-inspection tools (e.g. Microsoft Purview AI hub, Palo Alto AI Runtime Security)
Native evidence
Contributes discovery of AI use, prompt inspection and policy at the model boundary. Outside its view: what the tool call changed downstream, and the governed reversal of it.
SenseOn correlation
SenseOn joins the prompt to the identity, the change, and the rollback in one ledger. Decision enabled: can you undo what the agent did, and prove it?
AI security posture and prompt-inspection tools (e.g. Microsoft Purview AI hub, Palo Alto AI Runtime Security)
Contributes discovery of AI use, prompt inspection and policy at the model boundary. Outside its view: what the tool call changed downstream, and the governed reversal of it.
SenseOn joins the prompt to the identity, the change, and the rollback in one ledger. Decision enabled: can you undo what the agent did, and prove it?
SIEM export and search
Native evidence
Contributes retention and query over whatever provider logs you export into it. Outside its view: the 78% of MITRE ATT&CK techniques enterprise SIEMs have no detections for (CardinalOps 2025), the identity and endpoint context the export dropped, and any governed next step.
SenseOn correlation
SenseOn keeps provider evidence on your retention terms and correlates it with identity, endpoint, network and code evidence under one policy. Decision enabled: is this agent action chain complete, and what should be contained?
SIEM export and search
Contributes retention and query over whatever provider logs you export into it. Outside its view: the 78% of MITRE ATT&CK techniques enterprise SIEMs have no detections for (CardinalOps 2025), the identity and endpoint context the export dropped, and any governed next step.
SenseOn keeps provider evidence on your retention terms and correlates it with identity, endpoint, network and code evidence under one policy. Decision enabled: is this agent action chain complete, and what should be contained?
AWS Bedrock
Native evidence
Contributes IAM and model access controls, guardrails, CloudTrail API events, and optional invocation logging. Outside its view: cross-provider activity, endpoint and network context, and the downstream business actions an agent takes.
SenseOn correlation
SenseOn correlates Bedrock invocation and identity evidence with cloud, network, endpoint, and code telemetry. Decision enabled: was this agent action expected, and what should be contained?
AWS Bedrock
Contributes IAM and model access controls, guardrails, CloudTrail API events, and optional invocation logging. Outside its view: cross-provider activity, endpoint and network context, and the downstream business actions an agent takes.
SenseOn correlates Bedrock invocation and identity evidence with cloud, network, endpoint, and code telemetry. Decision enabled: was this agent action expected, and what should be contained?
Anthropic / Claude
Native evidence
Contributes SSO, SCIM, and role controls, plus enterprise audit logs covering user actions, system events, and data access. Outside its view: activity beyond the managed workspace, downstream tool effects, and cross-estate lineage.
SenseOn correlation
SenseOn correlates Claude activity with identity, endpoint, code, network, and SaaS evidence. Decision enabled: who used Claude, what data or code was involved, and what changed next?
Anthropic / Claude
Contributes SSO, SCIM, and role controls, plus enterprise audit logs covering user actions, system events, and data access. Outside its view: activity beyond the managed workspace, downstream tool effects, and cross-estate lineage.
SenseOn correlates Claude activity with identity, endpoint, code, network, and SaaS evidence. Decision enabled: who used Claude, what data or code was involved, and what changed next?
OpenAI / ChatGPT Enterprise
Native evidence
Contributes workspace identity and administration, compliance logs and APIs, and API data controls; platform retention is time-bounded unless exported. Outside its view: personal accounts, non-OpenAI tools, and downstream actions.
SenseOn correlation
SenseOn retains provider evidence on your terms and correlates it with the rest of the estate. Decision enabled: is this event isolated provider use, or part of a wider incident?
OpenAI / ChatGPT Enterprise
Contributes workspace identity and administration, compliance logs and APIs, and API data controls; platform retention is time-bounded unless exported. Outside its view: personal accounts, non-OpenAI tools, and downstream actions.
SenseOn retains provider evidence on your terms and correlates it with the rest of the estate. Decision enabled: is this event isolated provider use, or part of a wider incident?
Microsoft Copilot + Purview
Native evidence
Contributes tenant policy, identity, information protection, and DSPM and DLP audit of prompts and resources accessed when configured. Outside its view: non-Microsoft providers, endpoint and network behaviour, and code and runtime outcomes.
SenseOn correlation
SenseOn correlates Microsoft evidence with external AI, device, network, code, and business systems. Decision enabled: did protected data move through AI, and was the resulting action authorised?
Microsoft Copilot + Purview
Contributes tenant policy, identity, information protection, and DSPM and DLP audit of prompts and resources accessed when configured. Outside its view: non-Microsoft providers, endpoint and network behaviour, and code and runtime outcomes.
SenseOn correlates Microsoft evidence with external AI, device, network, code, and business systems. Decision enabled: did protected data move through AI, and was the resulting action authorised?
GitHub Copilot
Native evidence
Contributes enterprise AI policy and access controls, plus agent audit events with the initiating user, session, and action. Outside its view: local prompt and session detail, package and runtime behaviour, and cloud and SaaS effects.
SenseOn correlation
SenseOn correlates agent audit events with repo, package, CI/CD, endpoint, identity, and runtime evidence. Decision enabled: did an agent introduce or propagate risky code, dependency, or credential use?
GitHub Copilot
Contributes enterprise AI policy and access controls, plus agent audit events with the initiating user, session, and action. Outside its view: local prompt and session detail, package and runtime behaviour, and cloud and SaaS effects.
SenseOn correlates agent audit events with repo, package, CI/CD, endpoint, identity, and runtime evidence. Decision enabled: did an agent introduce or propagate risky code, dependency, or credential use?
THE SENSEON ARCHITECTURE
One system from evidence to governed action.
SenseOn is the complete system, not a middle stage between your tools. The Data Fabric feeds the Intelligence Fabric, the Agent Control Plane governs everything that acts on them, including the agents you did not build, and you choose how it is deployed. SenseOn correlates and governs these providers rather than replacing them: your Bedrock, Microsoft, GitHub and AI-posture investments stay in place.
Data Fabric
Ingests, shapes, and retains source-linked evidence from identity, endpoint, network, cloud, SaaS, AI providers, code, and business systems, on your retention terms, not each provider's default window.
Intelligence Fabric
Joins entities, sequences, detections, and prior context so investigations stay source-linked as the questions multiply, and so a model output and the change it caused sit on the same chain.
Horus and human authority
Horus coordinates bounded work while identity, scope, policy, approval, timeout, escalation and rollback stay explicit.
Decision Trace
Records the task, evidence, agent contribution and human decision in the Manage Case thread.
CORE SOLUTION
Provider logs, or the whole action chain.
Every provider audits its own workspace well. None of them can see the identity, endpoint, network, and code evidence sitting either side of the agent's tool call.
Provider logs only
Five consoles, five retention windows, five schemas
The prompt is visible; the change it caused downstream is not
Personal accounts and unmanaged tools leave no tenant record
Reconstruction means exporting logs and joining them by hand
Retention expires before the investigation asks the question
With SenseOn correlation
One evidence estate across providers, identity, endpoint, network, and code
Input, tool call, approval, action, and correction join into one chain
Provider evidence retained on your terms, not the platform's default
Containment and rollback run under policy you set, with named approvers
Task, evidence, agent contribution and human decision recorded in Decision Trace
ONE ACTION CHAIN
One AI action chain, hop by hop.
An example: an AI agent works a customer support ticket and issues a refund. Each hop leaves evidence in a different place, and the questions a regulator or board asks sit in the gaps between them. SenseOn joins input, tool call, approval, action and correction into one chain.
Hop
What each system records on its own
The evidence you need, and the question it answers
1. The agent reads the ticket
The AI provider logs the prompt and the model's reply.
The identity the agent ran as and the data it could read. Question: under whose authority did it act?
2. It calls a refund tool through MCP
The tool server may log the call, in its own console and format.
The OAuth grant behind the call and who approved that grant. Question: was this tool approved for this agent?
3. The tool changes the billing system
Billing logs a refund made by a service account.
The link from that service account back to the agent, the ticket and the prompt. Question: what did the agent actually change?
4. Who approved it?
Often nothing: any approval sat in a chat or a ticket comment.
A named approver, or the policy that allowed the action unattended. Question: who decided, and was that within policy?
5. Can it be reversed?
No single log holds the before-state or the undo step.
The before-state, the rollback step and who may run it. Question: can you undo what the agent did, and prove it?
HOW SENSEON WORKS A CASE
The same four steps run on every case.
The same governed workflow (Correlate, Fill in any Gaps, Investigate, Evidence Pack) runs on every case; it is walked through in full on AI for Security.
One real agent workflow, your own telemetry, a pass or fail reconstruction test, and an executive readout. Bounded scope, inspectable outputs, no rip-and-replace.
Week 1: Source map
Inventory AI providers, agent workflows, identities, and tools, and map the evidence each already produces. Output: a source coverage map and the reconstruction gaps.
Week 2: Bounded workflow
Select one agent workflow that crosses trust boundaries and define the chain-of-custody questions. Output: workflow scope, pass or fail criteria, and a response boundary.
Week 3: Evidence chain
Correlate provider, identity, endpoint, network, and code evidence for the selected workflow. Output: an inspectable chain from input to correction.
Week 4: Executive readout
Present the risks found, the remaining gaps, and the recommended governed scope. Output: a decision pack for funding and next steps.
ISO/IEC 27001:2022 certified (British Assessment Bureau, UKAS-accredited). World Economic Forum Technology Pioneer. Independently tested by SE Labs and AV-Comparatives. Rated 4.9/5 on Gartner Peer Insights.
Decision Trace inside Manage Case
Records the task, evidence, agent contribution and human decision
A structural guarantee, not a sampling rate. Chain-of-custody for NIS2, DORA, EU AI Act, and ISO 27001.
SE Labs AAA
Independent endpoint testing. AAA is the highest published rating band
AV-Comparatives A+
Independent endpoint benchmark. A+ is the top category
ISO/IEC 27001:2022
Certified by British Assessment Bureau (UKAS-accredited). Certificate at trust.senseon.io
Cyber Essentials Plus
Fully certified. UK government procurement gate passed
Last reviewed: August 2026. Every number on this page is published with its denominator, sample, and time window on proof points.
TESTIMONIAL
What holding the evidence chain looks like.
What changes for a security leader when telemetry, identity, and policy context finally connect.
“SenseOn gives us the visibility and confidence to treat security as an enabler of AI adoption, not a blocker. The ability to connect telemetry, identity and policy context is exactly the kind of foundation security leaders need as AI becomes part of everyday work.”
QUESTIONS AND NEXT READING
Direct answers, then the evidence behind them.
The first three cards are the visible answers emitted in FAQ schema. The remaining cards lead to deeper evidence.
