Patronus AI

Active, privately held. Raised a $50M Series B led by Greenfield Partners (announced June 25, 2026).

Sources 1, 2, 3, 4 · checked 2026-11-08

• Builds RL environments with tasks, graders, and simulated apps for alignment, coding, knowledge work, science, and computer use
• Expert pairwise reviews of agent transcripts, marking the better of two runs (preference data)
• Vendor states every environment is reviewed by domain experts and checked automatically before delivery
• Mostly custom builds calibrated to the buyer's model; samples usually sent first; data ownership set in each agreement
• Publishes research and benchmarks (for example GlobeBench and TRAIL)

Sources 1, 2, 3, 4 · checked 2026-11-08

• Labs training agents with RL that need environments and graders built to order
• Teams wanting expert preference judgments on long agent transcripts

Sources 1, 2, 3, 4 · checked 2026-11-08

Contact vendor (talk to us)

Sources 1, 2, 3, 4 · checked 2026-11-08

• SOC 2 (listed in site footer)
• HIPAA (listed in site footer)
• TISAX (listed in site footer)

Listed only as footer text with no report type or scope; ask the vendor for reports.

Listed as stated by the vendor. Scope can differ by product or subsidiary; request current reports.

Sources 1, 2, 3, 4 · checked 2026-11-08

• No public pricing
• The vendor says it uses SFT demonstrations only in its own research; client work is RL environments and pairwise reviews
• Training-gain figures on its site come from the vendor's own runs

Sources 1, 2, 3, 4 · checked 2026-11-08