Bridgewater
29.8%
Fewer mistakes across six financial filtering tasks.
Versus the best frontier model tested.
Read the studyLeading firms are building
internal AI.
Thomson Reuters
$40M
Invested to train its own proprietary LLM.
Thomson · August 2026
Read the announcementTravelers
30,000+
Employees with access to secure, in-house AI.
TravAI · January 2026
Read the announcementConfident doesn’t
mean correct.
Your answers need evidence.
Your data needs boundaries.
Why control matters
Confidence isn’t evidence.
Models can confidently invent facts. Private AI still needs sources, testing, and human review.
OpenAI’s researchSensitive work, shared.
Samsung restricted AI after employees submitted internal data to ChatGPT, TechCrunch reported.
Read the reportWho handles your prompts?
Figma lists external AI processors for prompts and outputs. Disclosed processing—not a breach.
Figma’s disclosurePrivate work. Public questions.
A dispute questioned the use of private Codex research. OpenAI says those prompts could not have influenced its system, including through training.
OpenAI’s statement