Description

The Australian Taxation Office is building and deploying AI models in-house without centralised policies and guidance for public servants, exposing it to legal and ethical risks, the auditor-general has warned. Human oversight has also been found lacking, with the agency not adequately monitoring its 14 newest AI models, some of which automatically flag potential non-compliance in tax affairs for review. It means that one of the government’s heaviest users of AI cannot be certain its models are free from bias or safe, the Australian National Audit Office (ANAO) said in a report labelling the tax agency’s AI governance only “partly effective”.

Summary

The Australian Taxation Office (ATO) is developing and deploying AI models in-house without centralized governance policies, leaving them vulnerable to legal and ethical risks, as highlighted by the Australian National Audit Office (ANAO). The lack of oversight and the absence of a comprehensive governance framework raises concerns about biases and safety in the AI models used to flag potential non-compliance in tax affairs. This situation underscores the necessity for robust AI governance frameworks to ensure ethical and safe AI usage, especially within significant governmental institutions. The ANAO's report criticizes the ATO's AI governance as only 'partly effective,' highlighting the urgent need for improved governance measures to mitigate potential harms in AI deployment.

Body