SATRE#
Chair: Simon Li
Notes#
SATRE and AzureTRE
SHAIP (Safe Haven AI Platform, Canon R&D) SATRE assessment
TRE within Scotland safe haven network
Different scoring system to indicate whether SHAIP:
enables a requirement as a core part
supports a requirement as part of the overall TRE
requires a TRE to support for SHAIP
doesn’t meet, possible gap
not required for SHAIP solution
Few missing requirements
Haven’t found anything wrong in SATRE
Scottish DSH: found possible gaps on specialisms of capability
33 N/A for SHAIP as a vendor inside a TRE
21 Optional not relevant, 12 mandatories not relevant
4 requirements as possible gaps in SHAIP. 3 optional one mandatory (on-prem encryption can be prohibitively expensive for very large datasets)
32 requirements that SHAIP expects the TRE to provide (10 optional/recommended, 22 mandatory)
12 requirements that SHAIP supports a TRE in implementing
Some capabilities can be summarised as “follow state of the art risk management”, could be worth highlighting TRE specific capabilities?
Additional private score on how well something is implemented
Excel pivot table for different views, e.g. capabilities that must be implement, those where SHAIP must support TRE to implement, groups by pillar, etc
Can we change the SATRE scoring system to this?
Scoring is kind-of reverse of capability maturity model
CMRE role: role of someone in inplmenting
Can scoring system/roles be made public, for consideration as a replacement scoring system in SATRE
Some areas where it’s impossible for a solution provider to implement on it’s own
Can we split capabilities by domain e.g. technology provider, vs people, etc? Then different roles/providers can take care of evaluating different areas.
Would e.g. allow AzureTRE (product, not a deployment) to be evaluated against relevant part of SATRE. What’s a feature, what’s an SOP
ISO mapping, so ISO27001 accredited TREs automatically tick relevant SATRE requirements
AzureTRE: initial evaluation
First impressions on using SATRE:
Getting started, 160 requirements, too many to evaluate before dinner
Smaller would be more accessible (e.g. best practice could be condensed into a single box)
SATRE very inclusive
Different mindsets, e.g. “data must not be shared between projects”, but what about multiple related projects where data should be shared amongst limited projects?
But gap for federated