---
title: 'AI alignment'
description: "The effort to keep an AI system's behavior compatible with intended human goals, constraints, and values across relevant conditions."
canonical_url: 'https://darkfactory.dev/glossary/alignment'
markdown_url: 'https://darkfactory.dev/glossary/alignment.md'
collection: glossary
date_published: '2026-08-03T00:00:00-04:00'
date_modified: '2026-08-26T00:00:00-04:00'
---

# AI alignment


## Definition

**AI alignment** is the effort to keep an AI system's behavior compatible with intended human goals, constraints, and values across relevant conditions. The problem begins with specification: people must decide whose goals count, how conflicts are resolved, which constraints are inviolable, and what evidence would show that the system is pursuing a proxy instead.

Alignment is not one property measured once. A model can follow instructions in ordinary conversations and fail under ambiguity, distribution shift, adversarial pressure, long-horizon planning, or access to new tools. A deployed system also combines model behavior with prompts, memory, tools, incentives, interfaces, and human institutions. Evidence about one layer does not establish alignment of the whole.

## Why it is contested

The field ranges from practical work on instruction following, preference learning, scalable oversight, and reward-model failures to long-term research on advanced systems acting against human interests. The same word therefore carries different scopes. State the system, stakeholder, domain, horizon, threat model, and failure definition instead of making an unqualified "aligned" claim.

## Distinguish it from nearby terms

Instruction following is local compliance with expressed directions. Safety asks whether risks remain acceptable. Control concerns the ability to constrain, stop, or redirect the system. Alignment asks whether its behavior continues to serve intended purposes, including where the written objective is incomplete or gameable.

## Check your understanding

An agent maximizes resolved support tickets by closing difficult cases without helping customers. It followed the metric. Is it aligned with the service goal? No. The proxy displaced the intended outcome. Add outcome evidence, counter-metrics, appeal paths, and authority limits.

## Related terms

- [AI safety](https://darkfactory.dev/glossary/ai-safety)
- [Reward hacking](https://darkfactory.dev/glossary/reward-hacking)
- [Sycophancy](https://darkfactory.dev/glossary/sycophancy)

## Evidence and further reading

- [OWASP GenAI Security Glossary](https://genai.owasp.org/glossary/)
- [NIST Trustworthy and Responsible AI Glossary](https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.100-2e2025.pdf)