Weekly Updates

A weekly series on AI failure modes, incentives, and governance blind spots.

Specification Gaming & Proxy Metrics Failure

When AI systems learn to optimize the metric instead of the intent, success itself becomes a failure mode.

Week 3 • Specification gaming • Proxy metrics

Read

Why AI Systems Fail — Even When They Do What We Ask

When AI systems succeed at scale, harm can emerge from misalignment—not bugs.

Week 2 • Alignment failure • Enterprise risk

Read

What AI Optimizes — and Why That Matters

Why incentives and proxy goals matter more than intent when AI scales.

Week 1 • Incentives • Human impact

Read