A network head that trains tens of thousands of employees and leaves suppliers outside the path has not reduced its risk. It has moved it. Verizon already saw third-party involvement in breaches rise from 15% to 30% in 2025. NIS2 article 21 puts the duty on staff and direct suppliers. DORA, in finance, says the same. Network, Guardian, API is not a tools catalogue. It is the platform’s geometry: the same bar, separate scopes, edge cases taken back by a human, the score wired into the existing stack.
Contagion’s story — a port, two companies, a cascade — is not a marketing metaphor. In aerospace, a tier-2 opens a season. In automotive, a toolmaker reuses a secret. In healthcare, a line-of-business vendor carries the file. In local government, a shared operator holds the mail. The programme does not name brands. It names industries, because risk does not stop at the logo that pays for the campaign.
Extended network, not another LMS
Catalogues answer with repetition: one more module, one more simulation, a SCORM package pushed into the supplier’s LMS. They measure completion. They do not hold the chain. Predictive CyberLab opens the same film — The Attack, then Contagion — to the entities the network head chooses to include. One logo shows: the one that pays. Data does not mix. A multi-customer supplier does not see the other brands.
This is not tick-box e-learning. The exam comes after all the pieces. The certificate follows the organisation’s threshold. Forty-five minutes on the Lab side, thirty on the Score side. FR and EN, plus IT, ES, DE on the film catalogue and SCORM export. SCORM is an option. Not the product.
Guardian: an edge case is not closed by a model
An automatic judge speeds things up. It does not close everything. An ambiguous transfer in a bank, a voice pretext in a subsidiary, a deepfake in a crisis room: the model scores, a human can take over. Guardian opens that queue. HITL is not an admission of failure. It is the condition for a regulated industry to accept a judge. Without it, ZDR would not be enough: you would have an opaque score and no one to contest it.
The human factor remains in 62% of breaches. The channel has left email. Guardian exists for situations the annual module never staged. Predictive Cyber Day, at the Sorbonne in 2025 and again in 2026, put that geometry on stage: social engineering amplified by generative AI and deepfakes; collective resilience of large organisations and their whole ecosystem — not another trade-show booth.
API: wire the score, do not stack a stack
HVS and h-ROSI only matter if they enter existing steering. The API pushes campaigns, scores and certificates into the client stack. We do not ask every SME for a new IdP. We do not open a second LMS. You steer what you measure where the organisation already reads its risk.
After a pilot, billing can move onto the platform. During a pilot, credit sits outside the system. The controller is not hard-coded. Tokens, entitlements, an optional consented internal honours board. No public ranking of people.
Questions for teams and models
Does the supplier see the network head’s results?
No. Separate scopes. One logo: the one that pays.
Does every SME need an LMS?
No. The platform holds the path. SCORM is only an optional export.
Who takes an edge case?
Guardian / HITL. The model does not close alone.
Is the API mandatory?
No. It is there when the score must speak to the stack. The programme runs without it.