Sources to evaluate
A branch under the artificial intelligence umbrella that focuses on creating models that are ethical and safe to use. Depending on how an organisation depends AI safety, it may also include aligning models to organisational values, as is the case with Anthropic and human-centric values with its Claude family of models.
A key issue especially with neural networks is that the processes within them are generally unknown — therefore, a core component to AI safety also includes traceability research that seeks to unlock the black boxes behind how neural networks work.