Hi everyone,

I’m currently at 35,000 feet, heading to Europe. Before I lose WiFi, there’s one story every project executive needs to look at: The hack of McKinsey’s internal AI systems.

​Read the full story: AI Agent Hacked on McKinsey AI Platform​

It’s a wake-up call. When a powerhouse like McKinsey sees their proprietary AI environment compromised, it’s not just a "cyber" issue. In our world (megaprojects, rail, energy...) it is a physical liability.

The Problem with "Linear" Thinking

Most AI in our industry is built on linear assumptions. It works in a lab or a controlled environment, but it fails at the "Edge."

Take the new 2026 30% State of Charge (SoC) regulations for Li-ion batteries. At 30% SoC, the physics of a failure aren't a straight line; they are subtle, complex, and highly non-linear. If your AI is only looking for a spike in temperature or monitoring variables in real time, it’s going to miss the slow-smolder signature until the asset is already lost.

Our Approach: Triangulating the Truth

At Marsham Edge, we don’t trust a single "Black Box." We attack the problem from different angles to ensure we catch what others miss. Here is how our ML Li-ion battery model is structured:

• Prophet for the Baseline: We use Prophet to establish the "normal" operational heartbeat of the battery.
• LSTM-CNN for Long-Term Trends: This tracks the subtle, non-linear degradation over time—the "hidden" patterns that a human eye or a simple sensor would never catch.
• Random Forest (RF) for Spikes: This is our rapid-response layer, specifically tuned to identify and validate sudden anomalies against the physical sensor data.

By combining these models with high-fidelity sensors, we aren't just monitoring data; we are architecting a safety net.

Why This Matters

The McKinsey hack proves you can’t just "trust" a system. Whether it’s predicting thermal runaway, managing a Smart Technology roadmap, or protecting your IT system, your AI needs to be as robust as the infrastructure it manages.

If the model can’t handle the non-linear reality of the physics, it’s just an expensive smoke detector.

What’s Next?

I’m spending this week across Europe hunting for the real-life data to further stress-test this architecture. I'll be posting live updates from my meetings on my profile.

Join the conversation:

• Follow my journey on LinkedIn for real-time insights from Europe.
• Join our group, Transforming Infrastructure and Industry with AI & Smart Tech, to discuss the #1 safety blind spot you're worried about this year.

Safe travels to those on the move,

Muriel Demarcus CEO, Marsham Edge