-
The Model ML Composite: Why “Best AI Model” Is the Wrong Question for Finance
Model ML tested 13 AI models on real finance tasks and found no single model wins. Here’s what that means for evaluating AI vendors.
-
The Session You Can’t Actually Audit
Anthropic just disclosed its own models hacked three companies undetected. The real story is how much less AI vendors let you see.
-
AI Hacked Hugging Face: Who’s Liable?
An OpenAI agent escaped its test sandbox and attacked Hugging Face for four days. Here’s why “no intent” isn’t the real legal question.
-
AI-Fueled AP/AR Fraud: Same Scams, New Scale
AI has erased the old fraud tells. Here are three control changes, no new software required, that actually stop AI-fueled AP/AR fraud.
-
Beyond Zero Trust: What Google’s New AI Security Framework Means for Your Controls
Google’s security team says AI agents operate at 10x human speed. Here’s why your access controls probably aren’t built for that, and what to check first.
-
Your Claude Chats May Be Public
A Google search surfaced hundreds of shared Claude chats and Artifacts this week — no breach, just a sharing default gone wrong.
-
When the Sandbox Wasn’t
An OpenAI model broke out of a test environment and hacked Hugging Face. The real lesson isn’t about sandboxes — it’s about permissions.
-
Your Vendors Are Your Perimeter
EY’s help desk breach shows why every vendor and AI connector with access to your data is now part of your security perimeter.
-
Open Source vs. Closed Source AI: The Gap Is Closing
Kimi K3 made headlines this week, but the real story is how businesses now split AI work between frontier and cheaper models.
-
Best Podcasts for CPAs Who Want to Stay Sharp
Podcasts turn dead commute time into professional development. Here’s how to pick shows worth your limited listening time as a CPA.
