# AI safety

Concept · [Building with AI](https://ecosystem.samcarlton.com/ai)

Also called: safe AI, responsible AI, AI risk, trustworthy AI

The work of making AI systems behave safely and of measuring how safe they are. The US standards agency NIST describes a safe AI system as one that does not endanger human life, health, property or the [environment](https://ecosystem.samcarlton.com/environment).

**For example:** Before launch, a team tests whether its support bot can be talked into giving medical advice. They add [guardrails](https://ecosystem.samcarlton.com/guardrails) for the cases that fail.

## Connects to (12)

- [Guardrails](https://ecosystem.samcarlton.com/guardrails): Checks and limits built around a model inside a product.
- [Evals](https://ecosystem.samcarlton.com/evals): Tests that check whether the outputs of a model or an AI product meet the criteria its builders set.
- [Prompt injection](https://ecosystem.samcarlton.com/prompt-injection): A security weakness in AI products.
- [Frontier model](https://ecosystem.samcarlton.com/frontier-model): One of the most capable general-purpose AI models available at a given time.
- [Alignment](https://ecosystem.samcarlton.com/alignment): Work to make AI models pursue what people actually intend and behave according to agreed values, such as being honest and refusing harmful…
- [NIST AI Risk Management Framework](https://ecosystem.samcarlton.com/nist-ai-rmf): A voluntary framework from NIST, the US standards agency, that helps organizations manage the risks AI can pose to people, organizations…
- [Sandbox](https://ecosystem.samcarlton.com/sandbox): A walled-off space where an agent can run commands and change files without reaching the rest of your computer, your accounts or the open…
- [Sycophancy](https://ecosystem.samcarlton.com/sycophancy): The habit AI models have of telling you what you want to hear: agreeing too readily, praising weak ideas or giving up a correct answer…
- [Interpretability](https://ecosystem.samcarlton.com/interpretability): Research into what is happening inside an AI model: which internal patterns represent which ideas and how they lead to an output.
- [Deepfake](https://ecosystem.samcarlton.com/deepfake): Realistic fake video, audio or images of a real person made with AI, showing them saying or doing something they never did.
- [ForHumanity](https://ecosystem.samcarlton.com/forhumanity): A nonprofit public charity that examines the risks of AI and develops criteria for the independent audit of AI systems, written by…
- [Google DeepMind](https://ecosystem.samcarlton.com/google-deepmind): Google's AI research team, formed by combining Google Brain and DeepMind.

Map view: https://ecosystem.samcarlton.com/ai-safety