SRI International Computer Science Laboratory
Assurance 2.0
Robin Bloomfield and John Rushby
Assurance cases augment earlier approaches to assurance such as
standards and guidelines by allowing more choice in selection of
techniques for ensuring and justifying safety and other critical
properties.
Building on this, Assurance 2.0 is an approach to the development
and presentation of assurance cases that is intended to make their
construction and assessment more straightforward, yet also more
rigorous. In fact, it is rigor that enables straightforwardness
because it reduces the "bewilderment of choice" and makes
assurance cases more systematic and predictable. Assurance 2.0
employs several ideas that are not in themselves new, but
integrates them in a way that we believe is coherent and
effective.
Our papers on Assurance 2.0 are listed below in reverse
chronological order. We suggest starting with either the
"Manifesto"
paper (broad but light on details) or the one from Cliff Jones'
Festschrift (more technical but also more narrowly focused). Look at
the 2-page
"Nutshell" (pdf) when you need a really high-level overview or
memory aid.
-
Quantifying Confidence in Assurance 2.0 Arguments
by Robin Bloomfield and John Rushby.
CSL Technical Report SRI-CSL-25-01, December 2025.
-
Assurance 2.0 in a Nutshell (PDF) by
Robin Bloomfield and John Rushby. CSL Technical Note October 2024.
-
Confidence in Assurance 2.0 Cases by
Robin Bloomfield and John Rushby. Also available as
arXiv 2409.10665.
Expanded version of a paper from
The Practice of Formal Methods: Essays in Honour of Cliff
Jones, Part I.
Springer LNCS 14780, pp. 1--23, Sept 2024.
Note that this paper and the report below are about confidence in logical
assessment;
the new report at the top is about probabilistic confidence assessment.
-
Assessing Confidence in Assurance 2.0 by
Robin Bloomfield and John
Rushby, Technical Report
SRI-CSL-2022-02, May 2022 and also available as
arXiv 2205.04522,
both revised May 2024
-
Defeaters and Eliminative
Argumentation in Assurance 2.0 by
Robin Bloomfield, Kate Netkachova, and John
Rushby, Technical Report
SRI-CSL-2024-01, May 2024;
also available as arXiv 2405.15800.
-
Assurance 2.0: A Manifesto by
Robin Bloomfield and John Rushby, keynote presentation at
29th Safety-Critical Systems Symposium
(SSS'21), February 2021.
Draft available as
arXiv 2004.10474
An overview of my other papers on
assurance
Clarissa/ASCE Tool support for Assurance 2.0
Honeywell, Adelard (now part of NCC Group), SRI, and UT Dallas
Clarissa is a tool suite that supports Assurance 2.0. It builds
on Adelard's ASCE and Clarissa/ASCE is available from Adelard: see the
CAE website.
Clarissa supports the construction and evaluation of assurance
cases using Assurance 2.0, and provides tools for logical and
probabilistic assessment, defeaters, and residual doubts.
It also has a synthesis assistant that can synthesize assurance
(sub)cases from templates provided in a theory.
In addition Clarissa has tools for exploring semantic properties. It can use
an LLM to translate claims into a logical representation, then perform
reasoning using Answer Set Programming with s(CASP) to examine
properties such as consistency, certain forms of well-formedness,
and completeness.
Applications
GDM Inability Case
In 2024, Google Deep Mind reported an investigation they performed to
see if LLMs could manifest "dangerous capabilities."
Their initial report is available as arXiv 2403.13793
with a detailed treatment at
arXiv 2505.01420.
This is organized as a safety case but does not follow any specifc method.
Recently,
Arcadia Impact (UK) Government Task Force
performed an external review
of this "GDM Inability Case" as they call it, structured using Assurance 2.0.
There's an informal blog
description and a full paper available as
arXiv 2604.21964.
Return to my bibliography page.
John Rushby (R u s h b y @ c s l . s r i . c o m)
