Almanac

Learning path

AI Safety Research: Labs, Models, and the People Behind It

AI safety isn't an abstract idea — it's being built, tested, and debated right now by a handful of organizations and the models they ship. This path traces the landscape from the foundational technique that makes AI assistants steerable, through the labs most invested in safety research, to the specific models where those ideas are put into practice.

No deep technical background required. Take the steps in order: each one adds a new layer to the picture.

Mixed level6 steps~42 min

6 steps

Begin →

  1. Anthropic

    Anthropic was founded specifically around AI safety concerns, making it the natural first lab to study in this context.

    Read →BeginnerIn-depth

  2. Claude

    Claude is Anthropic's primary model line and the direct product of its safety-first research agenda — the ideas above, in practice.

    Read →BeginnerIn-depth

  3. Claude Opus 4.6

    Claude Opus 4.6 is a concrete, current instance of where Anthropic's safety research has landed — useful for seeing how the principles translate to a deployed model.

    Read →BeginnerIn-depth

  4. OpenAI

    OpenAI pioneered many of the safety techniques in use today and offers a contrasting organizational approach to the same set of problems.

    Read →BeginnerIn-depth

  5. ChatGPT

    ChatGPT is where OpenAI's safety work meets the public at scale — a useful lens for understanding real-world alignment tradeoffs.

    Read →BeginnerIn-depth

  6. Google DeepMind

    Google DeepMind rounds out the picture as the third major lab with a serious safety research program, and a distinct research culture.

    Read →BeginnerIn-depth