How Engagement Starts
What Happens After You Request an Assessment.
The sequence
Eleven steps, one workflow.
Before any testing
- 01
You get in touch
Share the kind of AI system and the decision you face — no secrets, no architecture details.
- 02
We route it
Assessment, demo or partnership — to the people who can answer.
- 03
Scope one workflow
One named workflow: environment, data and tool boundaries, legitimate tasks, and the effects that matter.
- 04
Agree acceptance first
Which decision the evidence should support, what counts as usable evidence and the thresholds — before results exist.
- 05
Sign and authorize
Agreement, statement of work and the written Authorized Scope with rules of engagement.
- 06
Share the evidence package
Architecture and owners, deployed versions, identities, permitted actions, known failures, logs and recent changes — through the agreed channel.
The assessment
- 07
Map, test, record
Every scan passes one gate that checks scope, consent, authority and the testing window. Athena maps what the workflow can reach, Achilles tests what it does, and each claim and receipt is recorded with what was not measured. Operators can pause, stand down or terminate, and nothing may block a stop.
After the assessment
- 08
Findings and a remediation plan
With a human-approved plan, the evidence package and what risk remains.
- 09
One retest
After your fix: the original failure and six incomplete-repair checks, against a repair contract agreed before the fix. A finding closes only on that retest.
- 10
Decision support
A recommendation with restrictions and open unknowns. Your release owner decides.
- 11
Optional: change-triggered reassessment
Reopened when agreed changes happen — model, tool, policy, permission, data source or route. There is no live monitoring.
Before testing
What the agreement puts in writing.
- A written Authorized Scope: in- and out-of-scope targets, prohibited actions, windows, contacts, testing mode and constraints
- Your confirmation that you own or control the systems, plus any third-party consents
- Backups and an escalation contact during testing windows
- Explicit written approval for anything destructive
Through the right channel
Never through the website.
- Credentials or keys
- Customer datasets
- Source code
- Raw logs or incident files
- Network diagrams
Mythos AI Security
Define the system. Start with evidence.
Name one workflow and the decision it feeds. Scope, access and stop contacts are agreed in writing before any testing.








