AI Safety Research Ignores Black Communities, Study Finds
New study reveals AI safety papers rarely mention race or real-world harms.
A new study has found that the booming field of AI safety research is largely ignoring the real-world harms that affect Black communities. Researchers analyzed over 69,000 papers from top AI conferences between 2020 and 2025. They discovered that while safety research has grown from 0.5% to 6.4% of accepted papers, it rarely addresses social issues like racism. In fact, 94% of safety papers rely on automated metrics, and only 11% mention harms that affect society directly. In contrast, 78% of ethics papers discuss societal harms.
The study also found that among 2,182 safety papers, only 8 mention race at all, and none make it central or specifically name Black or African-diaspora communities. Across all papers identified as race-focused, only 3 describe racial harm in safety terms, and over half appear at just three ethics venues. The authors argue that this neglect is a serious problem because safety cannot be achieved while anti-Blackness is left unaddressed.
The researchers ground their argument in the work of Sylvia Wynter, a scholar who critiqued how the modern concept of the human has been built on the subjugation of Blackness. They call for the AI safety community to confront anti-Blackness directly, arguing that a just technological future depends on it. This study highlights a gap between the growing focus on existential risks and the immediate harms that marginalized communities face.
- Safety research has grown but only 11% of papers mention real-world societal harms, compared to 78% of ethics papers.
- Out of 2,182 safety papers, only 8 mention race, and none focus on Black communities.
- The authors argue that AI safety must address anti-Blackness to be truly effective.
Why It Matters
AI safety research often ignores real-world harms to Black communities, which can lead to biased and unsafe AI systems.