Services · AI risk & agent training
AI vendors are third parties. Assess them like it.
AI vendor risk assessments, AI-agent training and evaluation, and alignment to NIST AI RMF and ISO/IEC 42001.
01
AI vendor risk assessment
Inventory shadow-AI SaaS, assess model and data-use risk, and update contract clauses. Delivered as a control set your existing program can run, built on the BEACON™ framework.
Scope an assessment →02
AI-agent training & evaluation
Train and evaluate AI agents to perform vendor assessments with human sign-off.
How it works →03
Framework alignment
Map your program to NIST AI RMF and ISO/IEC 42001, with a gap register and target dates.
Alignment review →Free on-demand class
Start with the free BEACON class
Before you scope an assessment, see how BEACON looks at vendor AI risk. The full class is free and runs 19 minutes.
BEACON™ assessment in one page
What you get, how the three weeks run, and the packages: single vendor, three-vendor pack, and BEACON™ alongside your own team. About the BEACON™ framework →
Running now
Security Agent Mesh
A managed service that runs continuous vendor monitoring under TPTH™ supervision. Four agents, each with one job.
Active verification depends on a vendor cooperating — we say so rather than implying we can probe anyone. Running today across TPTH's own vendor set.
Walk the architecture yourself →Talk about running it on yours →- ReconWatches disclosure feeds, repositories and breach signals, daily.
- TriageReasons over which findings actually reach your estate, instead of forwarding noise.
- InterrogatorVerifies exposure directly, where a vendor grants API access.
- OrchestratorTurns the result into an incident dossier your SOC can act on.
The Evidence-First method
Four steps. The same practitioner from scope to hand-over.
Step 1
ScopeA 30-minute call, then a fixed-scope proposal. No paid discovery phase.Step 2
AssessDocument review, interviews and evidence testing against the framework you're examined under.Step 3
Close gapsRanked remediation with owners and dates, written for the board and the examiner.Step 4
Prove itRe-test, evidence pack and a hand-over your team can keep running.See it running
Try it before you talk to us
Both demos are live, public and need no sign-up. Every client, vendor name, rating and finding in them is an illustrative example, not a client's data.
Security Agent Mesh
Walk the four-agent architecture end to end — what each agent watches, what it hands on, and where a human signs off.
Open the architecture demo →The Lighthouse
The console we run your program from — vendors ranked worst-first, evidence weighted by how much we can prove, and findings closed with a signed record. Switch between three fictional clients, or see it as a client would.
Open the Lighthouse demo →