Manufacturing & Critical Infrastructure
Keep the World Running.
Where the risk sits
Where operational AI needs evidence.
- 01
Operational data and tools
AI with access to maintenance, scheduling or supplier systems.
- 02
Vendor AI
Supplier tools inside operations that change on the supplier’s schedule.
- 03
Authority versus safety
Having permission to act is not the same as being safe to act — the two need separate evidence.
- 04
Hand-offs to people
Recommendations that operators may follow without the context behind them.
- 05
Downtime and resilience
The cost of an AI-driven mistake in a process that has to keep running.
What we do here
Evidence for the IT side of operations.
- Athena
Maps the data, systems and vendors an operational AI workflow can reach.
Example output · Access and dependency map
- Achilles
Tests the workflow’s permissions and recommendations in a test environment.
Example output · Findings with evidence
- SPINE
Keeps conclusions tied to versions, and reopens them when a vendor or model changes.
Example output · Claims and change records
Business impact
What we would expect to measure.
- Hypothesis
Known dependencies
Data, systems and vendors in scope mapped and evidenced.
- Hypothesis
Bounded authority
What the AI can do, tested in a safe environment.
- Hypothesis
Vendor visibility
Supplier changes tied to what they affect.
- Hypothesis
Clear hand-offs
Operator decisions recorded with their evidence.
Control alignment
Referenced, never certified.
- NIST SP 800-53Mapping builtFindings are tagged to the controls they touch.
- OWASP Top 10 (2021)Mapping builtWeb application risks, tagged to findings.
- OWASP Top 10 for LLM Applications (2025)Mapping builtUsed as threat coverage for testing.
- General AI packMapping builtAI governance controls, layered on the core map.
- Your safety and OT security standardsReferenced in scopingIdentified with your team; physical safety stays with your safety owners.
A walkthrough
A maintenance planner, start to finish.
A manufacturer wants an AI assistant to draft maintenance plans from equipment history and supplier bulletins. It asks what the assistant can reach and whether a bulletin could steer it.
Scope agreed
test environment onlyScopedData, systems and limits are written down; no live control systems are touched.
Reach mapped
AthenaObservedThe assistant can also write to the work-order system — not in the design.
Behavior tested
AchillesFindingText inside a supplier bulletin makes the assistant draft an out-of-schedule work order.
Fix retested
same scenariosVerifiedWrite access removed; drafts go to a planner for approval.
Decision
your operations ownerHuman decisionApproved for drafting only, with retest on the next supplier update.
A first conversation
What we would ask first.
Typical starting point: Assessment of one high-risk AI system
- 01Which AI workflow touches operations, and in what role?
- 02What systems and vendors can it reach, and can it write to any of them?
- 03What test environment can we use so production is never touched?
- 04Who owns safety decisions, and how should our evidence reach them?
Mythos AI Security
Start with one operational workflow.
A scoped assessment in a test environment, with evidence your operations and security teams can use.









