A djinn six® product
ProbeSix™
Sooner or later a regulator, an auditor or a customer will ask how you know your AI is behaving.
A pass or fail from a security tool is not an answer. ProbeSix™ gives you one: evidence of how your AI behaves under attack, mapped to the framework they audit against, in a report you can put on the table.
What it costs to have no answer
The question does not go away when it goes unanswered. It gets asked again, by someone more senior.
A launch held
Sign-off waits while someone looks for evidence that was never produced.
A finding on the audit
“AI controls not evidenced” is a line nobody wants in the report to the board.
A law with a date on it
From December 2027 the EU AI Act requires high-risk AI systems to be tested and evidenced before deployment and kept that way through their lifecycle.
The quiet one
Not knowing yourself. Guardrails you have never tested are not evidence.
We were asked the same question
ProbeSix™ is a SaaS platform built by djinn six® after we were asked to provide LLM security assessments for an enterprise AI deployment. We reached for existing tools and found none of them built for the compliance depth the engagement required. ProbeSix™ is what we built instead.
It runs adversarial conversational attacks against your LLM endpoint, probing guardrails under sustained pressure and scores every finding against the exact clause of the framework you answer to: EU AI Act, ISO/IEC 42001, OWASP LLM Top 10, NIST AI RMF, MITRE ATLAS or DORA. You get a security score, a governance posture score and a findings report you can hand to an auditor.
How an engagement runs
Five steps and who does each one. Our consultants do the work the platform cannot, starting with finding the AI you run in the first place.
- 1
Map what you run
We map the AI you actually run, with you. You cannot secure what you do not know about.
Our consultants - 2
Set the target
Point ProbeSix™ at your endpoint and pick the framework you answer to. It assembles the attack plan.
Consultants and ProbeSix - 3
Attack it
Attack it like a real adversary: single questions, full conversations and everything in between.
ProbeSix - 4
Score it
Score every result against the exact clause of the framework you answer to.
ProbeSix - 5
Report and fix
A report you can hand to your board, your auditor or your regulator. Our consultants work the fixes alongside your engineers.
Consultants and ProbeSix
What you put on the table
A report you can hand to your board, your auditor or your regulator. Every finding cites the clause it maps to and the report opens with what was tested and what was not. Click a picture to look closer.
Why the answer holds up
See exactly what changed after a fix
Rerun the same assessment after remediation and the report sets the two runs side by side: what regressed, what improved and what stayed the same, test by test. The evidence is timestamped and usable in an audit.
Pressure that builds across a conversation
Most LLM security tools send single prompts. ProbeSix™ runs adversarial conversational attacks, holding sustained conversations with your model to test whether guardrails hold under real pressure. Guardrails that hold on the first prompt can erode across a session. If your controls degrade, you will see it in the report before your auditors do.
No new attack surface to justify
ProbeSix™ connects to AWS Bedrock, OpenAI, Anthropic, Meta Llama, Groq or anything that answers over an HTTPS JSON API. API keys are encrypted in AWS Secrets Manager and used only at scan time. On AWS Bedrock it assumes a role you grant in your own account so there is no key to hand over. In regulated industries a new credential management requirement is the objection that kills adoption. This removes it.
Next time they ask, hand them the report.
Go straight to probesix.ai or book a call and our consultants will run the first assessment with you.