AI Alignment.

The research and engineering effort to ensure AI systems pursue goals, follow instructions, and behave in ways that match the intentions of their developers and users. It belongs to the AI & ML Security category of cybersecurity.

The research and engineering effort to ensure AI systems pursue goals, follow instructions, and behave in ways that match the intentions of their developers and users.

Alignment work bridges machine-learning research, policy, and security. Techniques include supervised fine-tuning, reinforcement learning from human feedback (RLHF), reinforcement learning from AI feedback (RLAIF), constitutional AI, debate, scalable oversight, and interpretability. The field studies misalignment risks such as reward hacking, deceptive alignment, sycophancy, specification gaming, and emergent power-seeking behaviour in increasingly capable systems. Alignment is foundational to AI safety: a misaligned model that is otherwise secure can still cause harm because it pursues the wrong objective. Major labs (Anthropic, OpenAI, DeepMind) and institutions like the UK AI Security Institute publish alignment research, evaluations, and benchmarks that feed safety policies, red-team scenarios, and governance frameworks.

Defences for AI Alignment typically combine technical controls and operational practices, as detailed in the full definition above.

Common alternative names include: Value alignment, Model alignment.