A CISO’s guide · EU edition · for the person who signs off on autonomy
AI can triage alerts. But do you
trust it to act on yours?
Six criteria for adopting autonomy you can put your name on — and defend to the board, the regulator, and the NIS2 72-hour clock. Mapped to NIS2, DORA, and the EU AI Act.
The CISO’s Guide to Accountable AI SOC Operations
Triage is table stakes. Trust is the hard part — this guide is about the hard part.
Every AI SOC vendor can produce a verdict in seconds. None of that matters if you can’t see how it got there, contain it when it’s wrong, and defend it after the fact. This guide gives you the six criteria to judge any AI SOC platform — the ones that decide whether autonomy is something you can put your name on. Use them on us. Use them on anyone.
- 01Autonomy as a dialMatch the autonomy to the alert — not set-and-forget
- 02Hallucination containmentWhat the architecture does when it isn’t sure
- 03Reversible responseAutonomous action you can roll back in one click
- 04Investigation depthL2 reasoning on every alert, across your stack
- 05Memory you ownInstitutional knowledge that stays when people leave
- 06One audit trailThe record your regulator, board, and insurer can read
The real question
Every AI SOC gives a lightning-fast verdict.
You just need the right one.
The question that decides whether an AI SOC earns a place in yours is not whether it can produce a verdict. It’s whether you can trust it to investigate your environment, hand off when it isn’t sure, and prove what it did, why, and on what evidence.
A verdict you can’t interrogate isn’t an answer. Four questions decide whether an AI SOC is something you can actually run:
Can they see why it reached a verdict — or just the verdict?
One record when a regulator asks — or a weekend of stitching logs together after the fact?
AI strives to find the answer — which is why it hallucinates. What does the architecture do to contain that tendency?
Investigation can be probabilistic. The remediation that touches your environment and your users cannot be a coin flip.
Why trust is the whole game
A tool your team doesn’t trust
never leaves the pilot.
Your analysts will re-verify the AI’s work — for a year, maybe always. The stakes are too high not to. So the platform that earns trust isn’t the one that asks them to stop checking; it’s the one that lets them check everything — the full story of an alert, and why a verdict was reached — at a glance. AI isn’t perfect. Transparency is. What kills trust is not being able to see what the AI did, or why.
And when an analyst finds a crack, they correct it — and the AI never makes that mistake again. That’s how a pilot turns into production. Everything below is what earns that trust — and the criteria to demand of any AI SOC vendor, not just this one.
What this means for the person who signs off
You’re pushed from both directions at once.
A board and a mandate want AI in the SOC now. An examiner, an insurer, and a general counsel will ask — the morning after a material breach — who decided, on what basis, and can you prove it. You’re expected to move early and defend the move.
They want a defensible decision and a record they can actually read — not a black box.
They want one thing: more time on real threats. When the AI isn’t certain, it hands off to them — so they know they’re not being replaced. There’s still ample need for their context, their judgment, and their call.
Accountability is how you do both — adopt autonomous AI without it becoming your personal liability, and without your team feeling written out of the loop. Here’s what to demand.
Criterion 01 · Autonomy
Autonomy should be a dial — not set-and-forget.
A phishing email and a domain-admin compromise don’t deserve the same leash. In Morpheus you set the autonomy level to match the alert — and widen the dial as trust grows. Same reasoning engine at every level; only the amount of human sign-off changes.
Criterion 02 · Hallucination containment
A wrong answer should never reach a consequential action.
Most tools are built to triage everything, come hell or high water — so on the alerts they can’t resolve, they invent the evidence behind a confident verdict. Morpheus is built the other way. It triages what it can do confidently — with strict rules to hand off what it can’t.
In one SOC that might mean it handles 70% of alerts cleanly rather than claiming 95%. The rest is escalated on purpose, not closed on a guess. Reasoning is bounded; nothing destructive happens without a gate. The rule: certain, or it asks.
Grounded reasoning over your real telemetry, bounded so it can’t wander into invented facts — hard caps on iteration, cost, and tool scope.
Uncertain findings escalate to a human instead of auto-closing. A wrong answer can’t reach a destructive action.
A vendor claims 99% accuracy? In a SOC running 4,400 alerts a day, that’s 44 wrong calls every day — and the ones that hurt are the missed threats closed as benign. Ask for the containment story, not the accuracy number.
Criterion 03 · Reversible response
Autonomous response, in production today — on the right alerts.
Some alerts can’t wait for a human. A phishing blast mid-spread, a credential popping on an exposed host, a token exfiltrating — on those, every minute of delay is blast radius. Morpheus has already investigated the alert in full and orchestrated the response: isolate the host, disable the account, block the sender. It acts first and buys you time.
Because every action is logged and reversible, the call is never final — so your analyst gets the time to review what was done and why, on a contained incident instead of a live one. Disagree? Roll it back in one click. Fast action stops being a gamble.
And the action itself is deterministic — the same way, every time. Most tools bolt on and babysit a separate SOAR for that. Morpheus has a full deterministic SOAR built in — one engine, one bill, one audit trail.
Criterion 04 · Investigation depth
It investigates like a top analyst — on every alert, across your stack.
800+ integrations, so it works with the stack you already run — SIEM, EDR, identity, email, cloud. Attack Path Discovery correlates across all of them to find root cause before an analyst opens the alert.
As a Microsoft Intelligent Security Association (MISA) member, the Microsoft integration runs deep — native Defender, Sentinel, Entra, and Azure. Morpheus reads your Microsoft signal and adds the reasoning and response layer on top. It doesn’t replace what you own — it makes it more valuable.
Controlled benchmark: three multi-stage phishing scenarios mapped to MITRE ATT&CK, the same evidence available to both tools — a test of investigation depth, not a knock on Microsoft’s signal.
| On three multi-stage phishing attacks | D3 Morpheus 3 / 3 | MS Copilot 0 / 3 |
|---|---|---|
| Root cause | ✓Traced in all three | –Found in none |
| Kill chain | ✓Reconstructed 6–8+ stages | –Summarized, not connected |
| Correlation | ✓Email, endpoint, network, cloud | –Stayed within Defender’s data |
| Output | ✓Ready to contain & remediate | –Back to the starting line |
The two are designed to work together: Microsoft for detection, Morpheus for the verdict. Full methodology on request.
Member · deep Defender, Sentinel & Azure integration
Criterion 05 · Memory you own
The experience can’t walk out the door.
Morpheus absorbs your team’s local knowledge and special tactics — the VIPs, the contractor logins that look suspicious but aren’t, the scripts ten other tools keep flagging. When a senior analyst resigns, what they taught the system stays.
What it learns stays inside your environment: never pooled, never used to train anyone else, exportable if you leave. And it keeps its own integrations healthy — so your engineers spend their hours on hardening and hunting, not plumbing.
Scoped to your environment — never pooled, never used to train anyone else’s model, exportable if you ever leave.
It keeps its own connectors alive — engineers stop firefighting them mid-incident.
Every hour AI spends on maintenance is an hour your people spend maturing SecOps.
Criterion 06 · One audit trail
One record — the artifact your regulator,
board, and insurer actually read.
When the SEC, your auditor, or your cyber-insurer asks how a decision was made, the answer can’t be a weekend of stitching logs together. Many AI SOC tools spread the work across separate agents that each keep their own log — so the record gets reassembled by hand at audit time, exactly when you can least afford it. Morpheus runs one engine, on purpose, and produces one continuous record per incident, first alert to closeout, in open YAML.
That’s what holds up under NIS2 incident reporting and the DORA ICT-incident regime, gives you the traceability the EU AI Act expects of a high-risk AI system, satisfies the cyber-insurer at renewal, and — for those of you with US operations — meets SEC 8-K Item 1.05 and NYDFS Part 500 too. One provable, attributable record on demand is what shortens the disclosure clock and keeps you out of a finding.
For the European CISO
The six criteria, mapped to the regulation
you actually answer to.
Accountability isn’t abstract in Europe — it’s written into three overlapping regimes, and each one asks a question the six criteria above are built to answer. Here’s where they line up.
Extends incident reporting and risk-management duties across essential and important entities — with management-body accountability and an early-warning obligation within 24–72 hours.
For financial entities and their ICT providers: ICT-incident classification and reporting, resilience testing, and a demonstrable chain of who did what during an incident.
Where AI supports consequential security decisions, expects traceability, human oversight, and logging proportionate to the system’s risk — you must be able to show how the AI reached its call.
This mapping is guidance for evaluation, not legal advice. Obligations vary by member-state transposition, sector, and entity classification — confirm the specifics with your own counsel and DPO.
If that’s how you think about it
Then you should meet Morpheus.
An autonomous SOC with the guardrails and scale enterprise operations demand — plus everything in this guide. It investigates every alert, holds the autonomy dial you set, keeps one audit trail, runs your existing playbooks unchanged, and reads the SIEM, EDR, and cloud tools you already own.
Answer up to the board. Answer down to your team.
Accountable on every decision.
- See Morpheus run a live investigation, start to finish
- Get every question from this guide answered
- Bring your toughest alert and pressure-test it with our team
