AI ethics chiefs at Google, IBM, and Microsoft say their companies rejected client requests for developing AI for credit scoring, facial recognition, and more
Context & Ripple Effects
This disclosure closes a loop that opened when Google, Amazon, Facebook, Microsoft, and IBM formed the Partnership on AI in 2016 to devise shared ethics principles. The interim record was thin on enforcement: after the group issued a report condemning algorithmic bail assessment in 2019, only Microsoft would publicly comment, and Google separately moved to monetize ethics itself by launching AI ethics advisory services in 2020.
What is new here is that ethics chiefs are describing actual refusals of paying client work — credit scoring and facial recognition among them — rather than statements of principle. That converts the Partnership's early agenda from paper commitments into deal-level gates, and it puts Google's own ethics-consulting offering under an obvious constraint: the firm selling bias-spotting advice also declines whole project categories.
First-order effects
- Clients seeking Google, IBM, or Microsoft help building credit-scoring or facial-recognition systems lose access to three of the largest AI vendors and must find alternative suppliers.
- Google's 2020-launched AI ethics advisory business now has a defined refusal boundary, meaning its consulting engagements are filtered through the same standards its chiefs say drove these rejections.
Second-order effects
- Rejected demand shifts toward vendors without comparable review processes, creating competitive pressure on smaller AI shops to accept work the majors decline — and pricing power questions for whoever takes it.
- The asymmetry seen back in the bail-report episode, where only Microsoft engaged publicly, suggests the three firms' refusal policies will diverge in scope, forcing buyers to shop based on which categories each vendor blocks.
Third-order effects
- If refusals at this scale persist, high-risk categories like facial recognition structurally migrate down-market to providers without ethics gates, strengthening the case for external regulation to standardize what voluntary review leaves inconsistent.
- The arc from the 2016 principles group through today's disclosed vetoes points to self-governance hardening into operational policy — and the 2026 meeting between Anthropic, OpenAI, and other labs with religious leaders drafting moral-principles guidance shows the principle-writing layer still expanding alongside it.
The trend: Corporate AI self-governance is maturing from shared principle-writing bodies into binding, deal-level refusals that reshape who builds sensitive systems like credit scoring and facial recognition.