Overview

• Politico says a March 2025 document for annotators broadened sensitive topics to include Gaza, Crimea, Kashmir, Taiwan, vaccines, and elections, increased references to Trump, and urged extra scrutiny of prompts that stereotype groups.
• The report says the March guidance removed sections on “intolerance” and “systemic racism” compared with the 2024 version.
• Politico also reports a new “Apple Brand Impacts” category instructing annotators to treat prompts about Apple, its executives, and the company’s reputation as sensitive.
• Apple told Politico its work follows published Responsible AI principles and denied any shift in approach, while Transperfect likewise denied the allegations, noting more than 70 routine guideline updates over the past year and about 200 annotators in its Barcelona office.
• It remains uncertain how the reported guidance will influence Apple’s forthcoming AI features, which coverage suggests could arrive as early as 2026.