Excerpt – GovGPT FAQs
What are the security, privacy and accuracy
Released
considerations?
Security, privacy and trust are top considerations for GovGPT. Defined, publicly facing websites are
indexed to GovGPT’s model, which are then ringfenced as a technical barrier to containerise the
information GovGPT retrieves when forming its output. Indexing and ringfencing are mitigations for
hal ucinations that also support high accuracy outputs.
The model
does not col ect or store any data; and a pop-up disclaimer wil be mandatory on the site to
under
have users acknowledge the limitations of AI models and guidance to not input personal y identifiable
information.
An AI Impact Assessment has been conducted to ensure responsible development and governance of the
model.
GPT-4o comes with an advanced set of content filtering for themes such as hate, violence, harm and sex.
the
Additional y, GovGPT has been programmed to offer direction to 111 and/or Lifeline if it detects user
attempts to chat across these themes.
Official
GovGPT and the base model 4o have extensive security testing, threat monitoring and other measures
that can be viewed in detail at /www.openai.com/gpt-4o-system-card/.
Information
Act 1982
Document Outline
