AI, ML & Automation | Aligning Safety & Cybersecurity - Episode 6

Introduction

In this sixth episode of our Security and Risk Professional Insight series, we explored advancements and applications surrounding critical infrastructure protection, high-security environments, and the interplay of artificial intelligence (AI), machine learning (ML), and automation. This discussion included a focus on safety protocols and cybersecurity implications concerning these technologies.

Chris Cubbage, editor at My Security Media, hosted a distinguished panel comprising:

• Greg Sadler, CEO of Good Ancestors Policy
• Syy Go, Compliance Head at Advanced AI and member of the ISACA Emerging Trends Working Group
• Shannon Davis, Principal Security Strategist at Splunk Surge
• Dr. Mahendra Samarwickrama, Director at the Center for Sustainable AI
• Lana Tikhov, PhD candidate at the Australian Institute for Machine Learning

The session emphasized the significance of understanding AI's benefits, weaknesses, and the related regulatory and safety implications, as well as the necessity for comprehensive engagement with stakeholders.

Key Takeaways

Regulatory Trends in AI

Greg Sadler shared insights into the ongoing Australian government discussions regarding voluntary and mandatory regulations for AI safety. He emphasized the need for government action to address potential risks associated with AI capabilities. The evolving landscape of AI regulation aims to ensure that both high-risk usage (such as medicine) and general-purpose AI applications are supervised adequately.

The Role of AI in Cybersecurity

Syy Go focused on the current landscape of AI's integration into industries and the need for compliance policies and frameworks. He discussed surveys indicating that only 15% of organizations have effective AI policies, highlighting a gap in governance that must be addressed.

Security Challenges with Large Language Models

Shannon Davis elaborated on the security implications of large language models (LLMs) and the research conducted by Splunk Surge on securing these applications. With examples of prompt injections leading to unintended outcomes, Shannon stressed the importance of proper implementation of security practices around LLMs, pointing towards the OWASP Top Ten for LLM safety as a useful framework.

Human and AI Interaction Concerns

Dr. Mahendra Samarwickrama showcased how emerging technologies must navigate ethical considerations and human cognition. The discussion acknowledged the complexity of integrating AI into human decision-making processes and ensuring beneficial outcomes while maintaining ethical standards.

The Human Element in AI Safety

Lana Tikhov presented her research focusing on AI safety within medical applications, emphasizing the need to understand how AI interacts with human cognition. The implementation gap highlighted risks when AI systems produce errors or unexpected outputs, and the importance of fostering an environment where human oversight is vital.

Call for Collaboration

The panel concluded with the necessity for collaborative efforts toward developing robust frameworks for AI safety and regulation. Chris Cubbage called for a better understanding of how AI impacts both safety and cybersecurity sectors, reiterating the urgency of adaptive governance and responsible development to safeguard against the risks posed by AI technologies.

Keywords

AI, ML, Automation, Cybersecurity, Large Language Models, AI Safety, Human-Machine Interaction, Risk Management, Regulatory Frameworks, Compliance

FAQ

1. What is the main focus of Episode 6 in the Security and Risk Professional Insight series?

The episode primarily focuses on advancements in AI, ML, and automation, along with their implications on safety and cybersecurity in critical infrastructures.

2. Who were the panelists in this episode?

The episode featured a panel consisting of Greg Sadler (Good Ancestors Policy), Syy Go (Advanced AI), Shannon Davis (Splunk Surge), Dr. Mahendra Samarwickrama (Center for Sustainable AI), and Lana Tikhov (Australian Institute for Machine Learning).

3. What are the regulatory trends discussed in the episode?

The episode highlighted ongoing discussions in Australia surrounding both voluntary and mandatory regulations for AI safety, with a focus on addressing risks associated with high-risk AI applications.

4. How can organizations ensure compliance with AI policies?

Organizations can begin addressing compliance by adopting established AI policies, ensuring that they align with necessary regulatory frameworks, and actively engaging in risk assessments.

5. What security challenges are associated with large language models?

Security challenges primarily relate to issues such as prompt injection, data privacy, and ensuring secure implementation practices surrounding LLMs to prevent unintended, harmful outcomes.