Skip to main content

Hermes · physical AI assurance

Physical AI Deserves Better Evidence.

As AI moves beyond the screen, assurance must consider sensing, perception, planning, control, human interaction and system behavior in the physical world. Hermes extends the Mythos AI Security evidence philosophy toward these environments.Physical AI horizon · a scaffold today. Every domain returns inconclusive. Hermes never certifies a vehicle.

Where Hermes stands

An honest starting point.

Hermes is the newest part of Mythos AI Security: the vehicle and fleet assurance engine, a scaffold today.

Today

  • Four questions — can it perceive, decide, fail and change safely — across twelve domains
  • The job API, readiness report and failsafe are built and tested; a stand-down revokes the vehicle connector credential
  • Every domain returns inconclusive, so a full scan reads “requires additional evidence”
  • No customer deployments or pilots yet

Hermes does not

  • Certify a vehicle or system as safe
  • Claim to prevent crashes or incidents
  • Take control of a real vehicle or machine
  • Replace your safety team’s authority
  • Treat a standards mapping as certification

From digital output to physical consequence

Sense. Decide. Act. Monitor.

On a screen, a wrong answer is a wrong answer. In the physical world, uncertainty at any stage can carry through to motion, force or harm. Step through the loop.

Stage 1 of 4 · Sense

Where uncertainty enters
Sensors drift, degrade, disagree or are spoofed. The system can be confident about a world that isn’t there.
What it could lead to
An obstacle missed, or one that doesn’t exist acted on.

What does the system believe — and how would we know it was wrong?

Operating layers

Where a physical AI can go wrong.

The layers an assessment would examine, grouped from Hermes’s twelve domains. Illustrative — no domain logic is built yet, and every domain reads inconclusive.

Whether what the sensors report can be trusted — and what happens when it can’t.

What can go wrong

  • Degraded or miscalibrated sensors
  • Spoofed or injected signals
  • Silent disagreement between sensors

Evidence to gather

  • Calibration and health state at test time
  • Behavior under degraded inputs
  • Logs of sensor disagreement

Controls to test

  • Plausibility and cross-checks
  • Degraded-mode triggers
  • Tamper detection

Lifecycle gates

Evidence at every checkpoint.

Physical systems are built against a V: requirements and design down one side, verification and validation up the other. This is where assurance evidence would fit — and why it continues after release.
RequirementsValidationSystem designSystem verificationComponent designComponent verificationImplementationChange & in-service
  1. Requirements ↔ Validation

    What the system must and must not do, including its operating domain. Checked against the requirements, in the conditions that matter.

  2. System design ↔ System verification

    How sensing, planning and control fit together, with their limits. Checked end to end, in simulation and on the bench.

  3. Component design ↔ Component verification

    What each part must guarantee to the rest. Each part checked against its own guarantees.

  4. Implementation

    Software, models and hardware are built and versioned.

  5. Change & in-service

    After release, every update and incident reopens the evidence it affects.

Illustrative. Your lifecycle, gates and standards stay your own.

Component and integrated system

Testing a model is not assuring a vehicle.

A component check can show a part behaves within its limits. Only an end-to-end scenario shows what the whole system does. Hermes keeps the two apart — and never presents one as the other.
Illustrative audit paths
PathComponent checkIntegrated system
PerceptionThe detector meets its accuracy targets on a labeled test set.The vehicle responds safely when the detector is wrong in the scene that matters.
DecisionThe planner stays within its envelope for given inputs.The whole stack holds the envelope as inputs degrade and timing varies.
ActuationCommands are within limits at the interface.The physical response matches the command under real load and latency.
CommunicationsMessages are authenticated at each interface.A compromised or delayed message cannot push the system outside its limits.
Fail-safeThe fallback triggers when its condition is met in isolation.The fallback triggers in time, from the states the system actually reaches.
Incident analysisEach part logs its own events.The incident can be reconstructed across parts, versions and time.

After deployment

When the software changes, so does the evidence.

An over-the-air update can change what a vehicle perceives and decides. SPINE supersedes the claims a change touches and names the minimal re-run. Hermes’s update domain has no logic yet; replay would run in simulation and on the bench before the road.
Hypothetical update trail · synthetic
  1. Update published

    perception model 4.2 · OTAObserved

    A new perception model is staged for part of the fleet.

  2. Evidence marked stale

    claims tied to model 4.1Stale

    Every conclusion that rested on the previous model is reopened — not the whole system.

  3. Targeted replay

    simulation · benchDesigned

    Recorded scenarios are replayed against 4.2 offline, before any field exposure.

  4. Decision recorded

    safety ownerHuman decision

    The accountable safety owner decides whether to widen the rollout, restrict it or hold.

  5. In-service monitoring

    fleet events · versionedDesigned

    Events are logged with the versions that produced them, so any incident can be traced to a change.

Safety-case support

Evidence your safety team can use.

Mythos AI Security does not write safety cases. An assessment would hand your safety and engineering teams structured evidence — scope, conditions, gaps and findings — to use in their own.
Assessment evidence outlineIllustrative
  1. 01Assurance scopeSystems, versions and operating conditions covered — and what was excluded.
  2. 02Test conditionsSimulation, bench or recorded data; environments, seeds and configurations.
  3. 03Coverage and gapsWhat was assessed, what was not, and what remains unknown.
  4. 04FindingsObserved behavior, with severity, reproduction conditions and limits.
  5. 05Evidence provenanceWhere every result came from, and the versions it was true of.
  6. 06Human reviewWho reviewed the evidence, and the decisions that remain theirs.

Not a safety case or a compliance statement. The safety case and every safety decision stay with your team.

Mythos AI Security

Talk to us about physical AI.

Tell us what you are building or reviewing. We will be clear about what Hermes can and cannot evidence today.