Protopia AI Research
The Inference
Privacy Report
2026
We surveyed 350+ AI practitioners between November 2025 and May 2026 to find out where enterprise teams actually stand on using sensitive data with AI. Read the findings below, then take a 2-minute survey to benchmark your team against them.
350+
AI practitioners surveyed Nov 2025 to May 2026
89%
Say privacy has blocked or will block their AI projects
50%
Have not yet brought sensitive data into AI at all
Respondents from Oracle · NetApp · Disney · GoDaddy · Volkswagen · Cancer Research UK · Wells Fargo · Ericsson · Sony · Cisco · Prudential · F5 · Barco · NVIDIA
Key Findings
01
Enterprise AI is multi-model by default. Any privacy approach locked to one model or vendor is already behind the teams it needs to serve.
02
89% of practitioners say privacy has blocked or will block AI projects involving sensitive data. Teams are losing real use cases, not just anticipating problems.
03
Teams are already deploying privacy strategies like hardware isolation, masking, and confidential computing. But 76% of those teams still haven't made it to production with sensitive data. The current tools aren't closing the gap.
What the data shows
How enterprise teams use sensitive data with AI today: the models they run, where they deploy inference, and where privacy is blocking projects.
10+
Model Families
Running in production across enterprise environments. See which vendors lead, and how often teams run them side by side.
50%
Sensitive data maturity
Half of practitioners haven't brought sensitive data into AI at all. The other half is split between sandbox and production.
8
Data types in scope
From financial records to source code to PHI — privacy isn't a PII problem anymore.
89%
Privacy impact
Privacy has blocked or will block AI projects for nearly nine in ten practitioners.
Take the survey below to unlock the full report ↓