Enhancing AI Incident Data Sharing Using Cyber Threat Intelligence Feeds

This paper describes a new process and a prototype tool to enhance AI incident data-sharing methodologies. This addresses the increasing overlap between AI and cybersecurity incidents, and the need for standardised cybersecurity threat intelligence (CTI) feeds on AI incidents, indicators of compromise and threat actors. The prototype tool enables the Malware Information Sharing Platform (MISP) to consume semi-structured data from the Stanford AI, Algorithmic, and Automation Incidents and Controversies (AIAAIC) repository while ensuring scalable and efficient machine-based processing of AI incident data.We demonstrate how the tool successfully facilitated the integration of AIAAIC data into MISP, enhancing the ability to share and analyze AI incident data within the cybersecurity community or by using cybersecurity techniques such as incident time lining. These results show the feasibility of automated AI incident data serialization and integration of AI incident processing into established cybersecurity frameworks.This paper also identifies significant gaps and areas for improvement in the data models currently used to describe AI incidents. This work uncovered limitations within the AIAAIC data, particularly in the standardisation, machine-readability and completeness of incident descriptions; the variability in data formats and the lack of detailed, structured attributes. These hinder machine-based toolchain integration and automated analysis.