Algorithmic Failures Have No Home

LinkedIn Post 05: Algorithmic Failures Have No Home

Your building's HVAC AI model starts behaving erratically.

Energy consumption rises. Comfort complaints increase. Equipment runtime extends.

You contact the vendor. They investigate. They find a "model degradation issue" in the training data.

Your contract says: "Vendor shall continuously optimize building performance."

It says nothing about what happens when the model fails.

Is it a breach? An SLA violation? An act of God? Your contract doesn't specify.

The vendor's interpretation: "We're not liable. The contract just says we'll optimize. If the model fails, that's a technical issue, not a breach."

You hire a lawyer. The investigation costs more than the damages. You accept the degradation and move on.

This is the liability vacuum: when AI models fail, contracts are silent. Vendors face no accountability.

Solution: Establish explicit Service Level Agreements (SLAs) for algorithmic performance:

Energy consumption SLA: "Model shall maintain consumption within 105% of baseline"
Accuracy SLA: "Predictive alerts shall achieve 80% precision"
Failure SLA: "If model degrades SLAs for >14 days, vendor reimburses excess costs"

Algorithms should have explicit liability, just like equipment.