Who is accountable when AI makes a mistake?

This is one of the most important questions of our time. Currently, accountability falls on the humans who deploy AI — but as AI makes more autonomous decisions, accountability gaps grow wider.

Can we build accountable AI systems?

We can build transparent, auditable AI systems. But accountability requires moral agency — the ability to understand and suffer consequences. This is fundamentally a human capacity, not a technical feature.

February 9, 2026 · 5 min read · Juan C. Guerrero

Can AI Be Held Accountable?

An honest assessment of what AI can and can’t do — not hype, not panic.

AI Capability: Can’t Yet

AI can be audited. It cannot be held accountable. Accountability requires someone who suffers when they fail and is changed by that suffering.

What AI can do today

Here’s what AI systems can currently do in this area:

Log all decisions and reasoning
Provide audit trails
Maintain transparency about its processes
Accept corrections

These capabilities are real and improving. Anyone who dismisses them isn’t paying attention.

What’s still missing

Here’s what AI structurally cannot do — not “yet,” but by design:

Moral agency
The ability to suffer consequences
Reputation that can be damaged
Personal stakes in outcomes
The weight of knowing your decision affected someone's life

These aren’t just harder problems waiting to be solved. They require qualities that emerge from being alive, embodied, and mortal. In The Last Skill, I call this agency under consequence — the willingness to be the one who answers for the outcome.

Why this matters

Accountability isn't a log file. It's the knowledge that you will face the person you failed. In The Last Skill, this is called 'agency under consequence' — and it's what separates decision-making from processing.

The pattern is consistent: AI handles the predictable. Humans handle the consequential. The question isn’t whether AI can be held accountable at all — it’s whether the parts it can do are the parts that matter.

The bottom line

AI can be audited. It cannot be held accountable. Accountability requires someone who suffers when they fail and is changed by that suffering.

The distinction isn’t between humans and machines. It’s between capability and identity. AI is capable of many things. But the question “Can AI be held accountable?” is really asking: can it do the part that matters? And the part that matters is always the part that requires being human.

For the complete framework on what makes humans irreplaceable — including the four proofs of irreplaceability and why “agency under consequence” is the last skill — read The Last Skill: What AI Will Never Own by Juan C. Guerrero.

More from Anthropic Press: Will AI Replace Lawyers? · Will AI Replace Doctors? · 7 Skills AI Will Never Replace