The UK government’s Frontier AI Taskforce is building an
AI safety research
team that can evaluate risk at the frontier of AI. Doing great AI safety research does not mean
starting from scratch or working alone. As set out in its
first progress report
on 7 September 2023 the taskforce is working with leading
technical organisations including ARC Evals, RAND and Trail of
Bits.
Since then, the taskforce has partnered with a further three
leading technical organisations. These new contracts with Advai,
Gryphon Scientific and Faculty AI will tackle important
questions about how AI systems can improve human
capabilities in specialised fields and risks around current
safeguards. The findings of the research will be incorporated
into presentations and roundtable discussions with government
representatives, civil society groups, leading AI companies and experts in
research at the AI
Safety Summit in November.
Advai is a UK company focussed on enabling Simple, Safe, Secure
AIadoption. Their
technology and research focusses on identifying vulnerabilities
and limitations in AI to improve and defend these
systems. The Frontier AI Taskforce is working
with Advai to research vulnerabilities of frontier AI systems.
Faculty is an applied AI company, providing software,
consulting, and services. It has worked with the UK government
for nearly a decade, with its other public sector work including
partnering with the NHS to build the COVID-19 early warning
system, and with the Home Office to detect ISIS online
propaganda. The Frontier AI Taskforce is working with
Faculty AI to
identify to what degree LLMs can uplift a novice bad
actor’s capability, and how future systems may increase this
risk.
Gryphon Scientific is a physical and life sciences research and
consulting company with technical expertise in public health,
biodefense, and homeland security. They have experience in
working at the forefront of scientific advancement alongside
governments, including the US, and nations in the Middle East and
North Africa. Gryphon Scientific is working with the Frontier
AI Taskforce to
identify the potential for LLMs as a tool to drive rapid
progress in the life sciences.
Today’s announcement follows the progress report on the 07
September, where the Frontier AI Taskforce announced the
establishment of its expert advisory panel, the appointment of
two research directors, and several partnerships with
organisations.
Notes to editors
Ian Hogarth and DSIT
have responsibility to identify and address any actual, potential
or perceived personal or business interests which may conflict,
or may be perceived to conflict, with the Chair’s public duties.
Ian has agreed to a series of mitigations to manage potential
conflicts of interest. This agreement includes, for example,
divestments of personal holdings in companies building foundation
models or foundation model safety tools. Mitigations are being
put in place to address each of the potential conflicts with
effect from the start of his role.
In line with these processes the Taskforce Chair, Ian Hogarth had
no involvement in the awarding of the Faculty contract.
