Executive summary
Vulnerability discovery has outrun human remediation. A single Patch Tuesday can ship 163 CVEs, and AI-assisted patch-diffing collapses the window between disclosure and working exploit to hours. Yet the median enterprise still measures remediation in weeks. For critical-infrastructure operators — rail, power, water, pipeline, healthcare — the gap is not merely operational risk; it is a safety and regulatory exposure that conventional IT tooling is structurally unable to close.
Mythal is an autonomous vulnerability-remediation platform — a fix-control plane — purpose-built for these environments. It orchestrates twelve specialized AI agents to drive each finding through a deterministic state machine, from discovery to verified, evidence-backed closure. Where automation meets Operational Technology (OT) or Critical Cyber Systems (CCS), a dedicated OT Safety Officer agent holds explicit veto authority, and a deterministic Policy Guard enforces dual approval, maintenance windows, and tested rollback before any change is permitted to execute.
Mythal does not replace the human decision. It removes the toil around it — so the engineer's judgment is applied to the one change that matters, not the thousand that don't.
The result is a platform that compresses mean-time-to-remediate from weeks toward minutes for eligible findings, holds the line on safety for everything else, and produces compliance evidence mapped to TSA, NIST and IEC 62443 controls as a byproduct of normal operation rather than a quarterly fire drill.
The problem
Three forces have converged to make manual vulnerability remediation indefensible for critical-infrastructure operators.
Discovery has outrun remediation
The volume of disclosed vulnerabilities grows every quarter, and exploit development has been industrialized. Defenders no longer have the luxury of monthly patch cycles; they face an adversary who can weaponize a disclosure before the change ticket is even drafted.
Prioritization and change friction at scale
Knowing which of ten thousand open findings actually matters — and which can be safely automated — is a problem of evidence, not effort. Compounding it: change-control friction, rollback risk, fragmented compliance evidence scattered across spreadsheets and ticketing systems, and an inventory that drifts faster than anyone can map it.
OT changes the rules
In Operational Technology environments, a botched patch is not a help-desk ticket — it can stop a train, trip a substation, or halt a water-treatment process. Operators are right to veto any tool without an explicit, demonstrable safety model. This is precisely the gap IT-only remediation tools cannot cross.
What Mythal is
Mythal is a closed-loop, autonomous remediation platform — a fix-control plane that sits above an operator's existing scanners, configuration-management tooling, and OT systems. Rather than producing another queue of findings for humans to triage, Mythal takes ownership of each finding and drives it to resolution.
The control plane is built on three commitments. First, autonomy with brakes: agents reason, plan and execute, but a deterministic policy layer and the OT Safety Officer can stop or gate any action. Second, verification, not assumption: a fix is not "done" until Mythal has re-scanned, confirmed exploit-safety, and validated asset health — or rolled the change back. Third, evidence by default: every decision, message and step is recorded in a tamper-evident ledger and tagged to compliance controls as it happens.
A vulnerability scanner tells you what is wrong. A fix-control plane closes it — verifiably, safely, and with the paperwork already done. — Mythal design principle
How it works — the closed loop
Every finding in Mythal advances through an explicit state machine. The progression is deterministic and auditable; agents propose and act, but state only advances when entry conditions and policy gates are satisfied.
A finding is discovered from one of eight-plus scanners and normalized. It is enriched with threat intelligence, then prioritized by a composite risk score. The Remediation Planner produces a concrete plan, which enters awaiting-approval where the Policy Guard and — for OT/CCS — the OT Safety Officer decide whether it auto-applies, needs one approval, needs dual approval, or is denied. Approved plans move to executing via vetted integration drivers, then to verified by re-scan and health check, and finally to closed with full evidence attached.
The twelve agents
Mythal's intelligence is decomposed into twelve specialized agents coordinated by a Supervisor that owns the finite-state machine. Specialization keeps each agent's reasoning narrow, testable, and individually governable.
| Agent | Role |
|---|---|
| Supervisor | Orchestrator FSM; routes every finding through the state machine. Runs on claude-opus-4-7. |
| Scanner Liaison | Normalizes findings from 8+ scanners; deduplicates on asset_id + CVE. |
| Threat Intel | Correlates NVD, CISA KEV, EPSS, vendor PSIRTs, ICS-CERT, and MITRE ATT&CK for ICS. |
| Patch Hunter | Finds fixes and workarounds; assigns a reliability score from 0 to 1. |
| Impact Analyst | Joins CMDB + dependency graph into a business-impact profile. |
| Change Risk | Models failure rates, maintenance windows, canary, blast radius, and rollback. |
| OT Safety Officer VETO | Veto over OT/CCS changes; proposes compensating controls; enforces dual approval, window & tested rollback per NIST 800-82r3 & IEC 62443. Runs on claude-opus-4-7. |
| Remediation Planner | Builds the concrete plan: steps, order, approvals, rollback, verification. |
| Executor | Applies changes only via approved integration drivers; records per-step result and rollback. |
| Verifier | Re-scans, runs exploit-safety and health checks; closes the finding or rolls it back. |
| Compliance Reporter | Tags evidence to frameworks; exports PDF + JSON. |
| Inventory Insights | Proactive, CVE-independent: detects EOL, version sprawl, shadow IT, CCS-without-owner, identity hygiene. |
Integrations & drivers
Mythal acts on the estate exclusively through vetted integration drivers — each implementing a uniform apply_patch() / rollback() contract — so that execution is consistent, reversible, and confined to systems the operator has explicitly enabled. Findings are normalized from a broad set of IT and OT scanners.
Scanners (normalized)
Patch / Execution
Network / Security
Identity & Cloud
Operational Technology
Threat intelligence & feeds
Prioritization is only as good as the intelligence behind it. The Threat Intel agent continuously ingests authoritative feeds and grounds every enrichment in live data, with deterministic fallbacks so the platform remains operational even when an upstream source is unreachable.
- CISA KEV — live Known Exploited Vulnerabilities catalog, with an offline snapshot fallback and a one-hour cache to guarantee continuity.
- FIRST.org EPSS — exploit-prediction scoring to estimate the probability that a vulnerability will be exploited in the wild.
- Master CVE catalog sync — continuous reconciliation against the authoritative vulnerability record.
- ICS-specific sources — ICS-CERT advisories, vendor PSIRTs, and MITRE ATT&CK for ICS for OT-aware context.
Risk scoring & prioritization
Mythal computes a composite risk score per finding rather than relying on CVSS alone. The score blends severity, real-world exploitation signal, the dependability of an available fix, and the operational context of the affected asset — and resolves directly into a policy-gate decision.
- CVSS base severity, combined with EPSS exploit-prediction and KEV status for known active exploitation.
- Ransomware association uplift for vulnerabilities tied to known campaigns.
- Patch-reliability band from the Patch Hunter, so a fragile fix is never auto-applied.
- Business-impact and change-risk profiles from the Impact Analyst and Change Risk agents.
The composite resolves to one of four decisions — auto_apply, single_approval, dual_approval, or deny — subject always to OT override.
OT safety & the Policy Guard
This is where Mythal departs from every IT-centric remediation tool. Operational Technology is read-only by default. No change reaches an OT or CCS asset without clearing both a deterministic policy layer and the OT Safety Officer agent, which holds explicit veto authority and can substitute compensating controls when a direct patch is unsafe.
The Policy Guard is a deterministic rule engine — extensible via OPA — so its decisions are predictable, testable, and never delegated to a probabilistic model:
SG-POL-001CCS changes require dual approval, a maintenance window, and a valid rollback plan.
SG-POL-002OT-zone changes require dual approval — security and OT operations.
SG-POL-003IT changes auto-apply when criticality ≤ Medium, patch reliability ≥ 0.85, a canary peer is present, the window is open, and rollback is valid.
SG-POL-004Default to single IT approval where no stronger rule applies.
SG-POL-006A CCS change without a rollback plan is denied.
SG-POL-007Any change during a blackout window is denied.
OT zones are modeled and segmented in line with IEC 62443 and NIST 800-82r3. The OT Safety Officer enforces dual approval, maintenance-window adherence, and a tested rollback for every change touching the plant.
Compliance & evidence
Because every agent message, decision, approval, and execution step is recorded as it happens, compliance evidence is a byproduct of normal operation rather than a separate effort. The Compliance Reporter tags evidence to control frameworks and exports auditor-ready PDF and JSON in seconds.
An auditor asks "show me how this CCS vulnerability was remediated," and Mythal returns the full reasoning trace, the approvals, the policy decision, the execution log, and the verification result — already mapped to the relevant control.
Architecture & technology
Mythal is engineered for production critical-infrastructure environments, with a deterministic rule-engine fallback that lets the platform run with no external model dependency for CI and demo safety.
- Backend — FastAPI on Python 3.11.
- State & bus — PostgreSQL 16 for durable state; Redis 7 as the agent message bus.
- Reasoning — Anthropic (claude-opus-4-7 / claude-sonnet-4-6) or OpenAI, with a deterministic rule-engine fallback that has no external dependencies.
- Console — Next.js 15 / React / TypeScript, with Recharts, Reactflow, and SWR.
- Deployment — Docker Compose; engine packaged as
sentinelgrid-api.
The console
The Mythal console gives security and OT teams a single command center over the entire remediation lifecycle. Its pages are designed for the realities of a critical-infrastructure operation — including a dedicated OT Operations view and a drag-to-approve plan board.
Operate
Estate & OT
Govern
Connect
Outcomes & evaluation
Mythal is evaluated against operational KPIs that map directly to executive and regulatory concerns. The targets below frame the value the platform is built to deliver.
Security & governance
Mythal is built to be trusted with production change in safety-critical environments, which demands a higher bar than typical SaaS automation.
- Tamper-evident ledger — every inter-agent message is signed with HMAC-SHA256 and written to an append-only log.
- Prompt-injection defense — external content is wrapped in untrusted tags and screened by a pre-flight classifier before any agent reasons over it.
- Write-ahead message log — actions are durably recorded before execution, ensuring recoverability and a complete audit trail.
- Full reasoning trace — every finding carries an end-to-end record of how each decision was reached.
- Deterministic fallback — a rule engine with no external dependencies keeps the platform safe and operable for CI and demos.
- OT read-only default & veto — the plant is protected by design, not by configuration.