AI Glossary

AI safety

The work of keeping AI systems from causing harm, whether through malicious use, malfunctions or wider risks to society. It spans research, testing and monitoring of AI systems, and the public policy that governs them.

See also

  • Alignment

    Making AI systems behave in line with human intentions and values, including in situations their makers did not foresee.

  • Red teaming

    Deliberately attacking an AI system the way an adversary would, to find harmful outputs, security holes or failures, before or after it is released.

  • Interpretability

    Research into what happens inside a model, aiming to explain how it gets from its input to its output.

  • Guardrail

    A rule or check around a model that blocks or corrects unwanted input or output, such as a filter for unsafe content or a limit on which actions an agent may take.

Source: Bengio et al., International AI Safety Report (2025) (opens in a new tab)

Cite this term.

CC BY 4.0, attribution required.

This term

Orbyt Labs, AI safety (AI Glossary), Oct 18, 2026. https://www.orbytlabs.ai/orbyt-collective/glossary/ai-safety

The dataset (JSON)

Orbyt Labs Glossaries (Oct 19, 2026), https://www.orbytlabs.ai/glossary#dataset

Last updated .