University of Manchester researchers found that AI engineers recognise ethical risks in the systems they build. They lack the authority, incentives and support to act on them. The researchers call this "ethical awareness without ethical agency". Interviews covered software engineers in finance, AI research and semiconductor manufacturing. Many described safeguards they wanted to add but could not. The work was presented at the 10th Data for Policy Conference in Barcelona. It landed days after Jacob Coxon resigned from Anthropic in a public post on X. Coxon said colleagues privately fear AI could kill everyone by the end of the decade. He accused Anthropic and OpenAI of racing to self-improving systems. HR Dive notes that a written AI ethics policy is one guard against this culture.

• conference held Sept. 8-10, 2026