Apple denies changing AI training rules after Trump election

Apple has strongly denied a Politico report claiming the company modified its AI training guidelines following Donald Trump’s election, specifically around topics like diversity, equity and inclusion (DEI), vaccines, and Trump-related content. The denial comes amid broader industry scrutiny over how tech companies handle politically sensitive topics in AI development.

What Politico claimed: The publication reviewed internal documents showing Apple updated its AI training guidelines in March 2025, allegedly making significant changes to how its models handle sensitive political topics.
• The report claimed sections on “intolerance” and “systemic racism” were removed from training materials.
• Topics like DEI policies, Gaza, Crimea, Kashmir, Taiwan, vaccines, and elections were reportedly added to a “sensitive” topics list.
• The March document allegedly included more detailed guidance on handling Trump-related issues, though with no changes to the model’s ultimate behavior regarding the former president.

How the training process works: Apple contracts with Transperfect, a translation services company, for AI data annotation work at their Barcelona office with about 200 employees.
• Data annotators label and categorize raw data, evaluate AI-generated outputs, and flag issues like bias or safety risks.
• Their structured feedback helps researchers fine-tune and retrain AI models.
• This annotation process is considered one of the most critical steps in AI training across the tech industry.

In plain English: Think of data annotators like editors who review and grade AI responses, teaching the system what constitutes good or problematic answers. Their feedback helps the AI learn to handle different topics more appropriately, much like how a teacher’s corrections help students improve their work.

Apple’s response: An Apple spokesperson categorically denied changing the company’s overall AI policy approach, though didn’t specifically address the individual changes mentioned in the report.
• “Apple Intelligence is grounded in our Responsible AI principles, which guide every step from training to evaluation. Claims that we’ve shifted this approach or policy are completely false,” the spokesperson said.
• The company emphasized it trains its own models and works with third-party vendors using structured topics, including sensitive ones, with regular updates to improve performance.

Transperfect also pushes back: Co-CEO Phil Shawe denied the claims but provided a somewhat ambiguous response about the company’s work processes.
• “These claims are completely false, and we deny them in the strongest possible terms,” Shawe stated.
• He noted the company regularly receives updated guidelines, with more than 70 updates over the past year, but claimed “none of these changed any policy, which has remained consistent.”
• Politico noted Shawe’s statement didn’t specify which particular claims he was refuting.

Why this matters: The controversy highlights the ongoing tension tech companies face in developing AI systems that handle politically charged topics while maintaining neutrality and avoiding bias accusations from multiple sides of the political spectrum.