AI Will Never Be Ethical or Safe

The argument

It has been reported that Jens Meiert, an engineering lead and self-described guerrilla philosopher, argues in a recent blog post that AI can never be entirely ethical or safe. Why? Because ethics and safety hinge on context and intent — things machines can’t reliably know. Simple examples make the point: “how to pull oxygen out of a room” is harmless in a chemistry class and catastrophic in a nursery; learning to use a firearm is legitimate for hunting and lethal with the wrong intent. Short version: without context and honest intent, a perfect rulebook is a hollow notion.

Industry trade-offs laid bare

Meiert points to real-world attempts to patch the problem — Anthropic’s “constitution” is singled out — and notes that companies already make judgment calls. It has been reported that Anthropic’s guidance treats some answers as “probably fine,” balancing harm mitigation and usefulness. But Meiert calls that naiveté: most users don’t declare malicious intent, and humans rarely provide full context to each other. The emotional crux here is trust — we build systems on social contracts that are fragile, and AI inherits that fragility. Is it comforting? Not really.

So what now?

The takeaway isn’t doom-mongering: Meiert doesn’t claim safety frameworks are worthless; he says they’re incomplete by design. AI remains a powerful tool that will be used for good and ill, and policy, product design, and public expectations need to reflect that limit. Expect guardrails, not guard angels. In short: demand better systems, but don’t fool yourself into thinking a model can read minds — or replace the messy human work of asking the right questions.

Sources: meiert.com, Hacker News