Microsoft launches global AI red team alliance for strong safety

Microsoft has launched the External Red Team Alliance (EXTRA) to enhance AI security through global collaboration. The initiative addresses the limitations of internal AI safety testing by incorporating diverse expertise from academia and independent researchers across six continents. EXTRA consists of two components: a global academic network that provides funding to 18 universities for AI safety research, and a distributed operational collaboration network that engages specialists in red teaming.

The aim is to improve methodologies in AI safety evaluation and ensure that research is independent and unrestricted, thereby fostering stronger connections between academia and operational teams. This proactive approach is deemed critical as AI capabilities continue to evolve, necessitating broader engagement to identify and mitigate emerging risks.