Securely enabling AI, grounded in risk

Understand where AI risk is introduced and exposed — then put in place what reduces it.

Built on the CoSAI Risk Map that succeeded Google’s SAIF, with MITRE D3FEND and ATLAS capabilities and the products that deliver them.

• IntroducedWhere the risk enters the system
• ExposedWhere it becomes visible or does damage
• MitigatedWhere controls can stop it

From risk to enablement

1. 1RisksStep through every risk and watch it light up where it is introduced, exposed and mitigated.36 risks
2. 2CapabilitiesThe MITRE D3FEND and ATLAS countermeasures that reduce each risk, grouped under the CoSAI controls they deliver.76 capabilities
3. 3ArchitecturesHow each class of AI application is built and how data moves through it, with every capability pinned where it goes.18 architectures
4. 4EnableOn each architecture: the foundations, the surface baseline and the use case's own additions, then how far each product covers them.52 products

See it happen: 20 real 2025–26 incidents walked step by step across their architectures, every step sourced.

Reference

• Risks36
• Controls35
• Components23
• Personas8
• Frameworks8

Where the data comes from

The taxonomy is CoSAI’s; the composition is SAIF’s. Where the two disagree, the choice is declared on the component it affects. Source and full audit on GitHub ↗