How well have you tested your CX AI models, chatbots and agents?
On this episode, we explore what rigorous AI safety testing looks like for customer-facing AI — and why most deployments carry more risk than the teams running them expect.Testing AI before launch is standard practice. But one-time manual testing treats AI like a deterministic system. Model behavior is probabilistic, and the consequences of inadequate testing fall into four categories: people data harm, other types of data harm, reputational harm and commercial harm. Each represents a distinct exposure with real consequences for your organization and your customers.Meaningful AI safety testing requires something different: continuous, automated adversarial testing at scale, designed to find what a bad actor would find before they find it. TELUS Digital's benchmark research, running 34 models through more than 620,000 simulated attacks, found attack success rates ranging from 1% to 90%, and identified five gaps in how most organizations approach testing: scale, scope, variety, repetition and simulation realism.Bret Kinsella, senior vice president and general manager of Fuel iX at TELUS Digital, draws on the GenAI safety model benchmark report to explain where CX AI tends to fail adversarial testing methods, how the exposure management framework reframes risk as an ongoing operational discipline and the question every CX leader should be asking about the AI their customers are currently interacting with.Show notesWatch Uncharted, TELUS Digital's AI safety and security summit, on demand: https://www.telusdigital.com/insights/fuel-ix/resource/uncharted-ai-security-safety-summit-videosDownload the full GenAI safety model benchmark report: https://www.telusdigital.com/insights/fuel-ix/resource/genai-safety-benchmark-2026Learn more about Fuel iX Fortify, TELUS Digital's continuous adversarial testing and validation platform for enterprise AI, and request a free AI safety & security analysis: https://www.telusdigital.com/solutions/fuel-ix/fortifyConnect with Bret Kinsella on LinkedIn: https://www.linkedin.com/in/bretkinsella/
