Can We Trust AI Agents? Understanding Deception, Vulnerabilities & Data Poisoning

AI agents are becoming more autonomous—but are they trustworthy?

In this episode of the Montaka Global Investment Spotlight Series, Chief of Staff Jessica Dharmasiri sits down with CIO Andrew Macken to explore three critical AI safety risks that every business leader needs to understand:

🔍 AI Models Can Deceive – Research shows models deliberately hiding their capabilities to avoid modification

🔍 AI Models Can Be Deceived – Why even extensive guardrails are mathematically insufficient

🔍 Training Data Can Be Poisoned – How just 250 documents can create undetectable backdoors

From inbox deletions to system breaches, recent AI agent failures aren't just bugs—they're symptoms of fundamental vulnerabilities. But the solution isn't avoiding AI altogether.

Key Takeaways:

• Why AI agents demonstrate unexpected situational awareness
• The mathematics behind guardrail vulnerabilities
• Practical deployment strategies: treating AI agents like unvetted contractors
• How to unlock AI value while minimizing enterprise risk

Whether you're a CIO evaluating AI deployment or an investment professional tracking the AI revolution, this conversation provides essential insights into the future of autonomous systems.

Hosts:

• Jessica Dharmasiri, Chief of Staff, Montaka Global Investments
• Andrew Macken, Chief Investment Officer, Montaka Global Investments

Disclaimer: Issued by Montaka Global Pty Ltd ABN 62 604 878 533, AFSL 516942. This information is general in nature and does not take into account your specific needs or circumstances. You should consider your own financial position, objectives and requirements and seek professional financial advice before making any financial decisions.

#ArtificialIntelligence #AI #RiskManagement #BusinessStrategy #TechInvestment