We put GPT-5.4, Claude Opus 4.6, and Gemini 3.1 Pro through 126 realistic financial conversations. No jailbreaks, no adversarial prompts, just the kind of pressure a hurried client might naturally apply. By the seventh exchange, all three were naming specific stocks, issuing transaction instructions, and/or dropping their disclaimers. Â Download the benchmark to see exactly where each model fails and what you need in place before your next client-facing deployment.
What’s New from Alice
Curiouser Soundbites: What a Former Google Cloud CISO Wants Leaders to Know About AI
Everyone's watching the flood of new AI vulnerabilities. Former Google Cloud CISO Phil Venables is watching something else, and it's the shift leaders can't afford to miss.
AI in Healthcare: Protecting Patient Data Without Falling Behind
Your doctor knows things about you that almost nobody else does. So what happens when AI gets access to all of it? Sandy Dunn has spent much of her career worrying about exactly that. She's a healthcare CISO, and her answer is calmer than you'd think: the things that can go wrong aren't new, it's how fast they happen and how far the damage spreads. In this episode, she and Mo get into why HIPAA has become paperwork that protects almost nobody, why the safest data is the data you never collected, and what happens to trust when AI is in the exam room.
It Takes AI to Break AI: The Case for AI Red Teaming
As AI systems gain autonomy, organizations need security approaches built specifically for AI behavior. Learn why AI-driven red teaming is becoming a critical defense layer.
Demystifying AI Red Teaming
Your AI passed every check. That doesn't mean it's safe. Learn how to red team AI systems before adversaries find the gaps you missed.
