Artificial intelligence (AI) tools have become extremely popular since the November 2022 release of ChatGPT from OpenAI, followed by the subsequent release of similar chatbots by other companies. However, questions about how to evaluate and secure these AI systems have existed for years. Microsoft is starting to share details about its AI red team, which has been working behind the scenes since 2018.
A ‘red team’ is a group which is paid by an organization to mimic both physical and digital attacks against the organization that hires it; and, subsequently to provide reports to that organization, identifying any weaknesses in the organization’s security in order to modify or eliminate those weaknesses.
Over the past five years, Microsoft’s AI red team has grown from an experiment into a team of AI and machine learning experts, cybersecurity researchers, and social engineers. The team aims to publish its findings in a way that is accessible to people and organizations, having no specialized AI knowledge. They have concluded that AI security differs from traditional digital defense and requires a unique approach.
Microsoft’s AI red team’s addresses both traditional system security and broader AI system weaknesses and failures, such as a rogue AI systems generating misleading, offensive and prohibited content. Since its inception, Microsoft’s team has released numerous security tools and frameworks to the public.
The red team’s findings in its early years proved the importance of having an AI red team. Currently, the AI red team is more in demand than ever because AI platforms are facing increased attacks from a wider range of assailants, and not just the most highly resourced ones. Even casual users or teenagers can target AI systems and bring them down.
Microsoft’s AI red team focuses on anticipating future attack trends and strategies, and working collaboratively within the company to fix the vulnerabilities it discovers. Furthermore, that mission includes addressing both AI and security failures, and making those findings public.
The whytry.ai article you just read is a brief synopsis; the original article can be found here: Read the Full Article…
