AI Risks: Privacy & Fraud Top Concerns in New Study

A new study of 34 AI models reveals that privacy, fraud, and cybersecurity are the riskiest categories. TELUS Digital conducted over 620,000 adversarial tests, finding vulnerability rates from 1.3% to 93%, meaning no model is fully secure. What's interesting is that reasoning models were the hardest to exploit, while smaller models showed the most vulnerability. The study highlights the need for continuous, automated testing to protect enterprise AI applications. TELUS points to its Fuel iX Fortify platform as a key tool for this. The bottom line: as AI use grows, understanding and mitigating these security risks becomes increasingly important for everyone.