#security · Priya N.
94%
We ran Type II last year, I still have the evidence checklist and the auditor's punch list…
Give your teams AI that gets work done
Find knowledge across your company
Protect sensitive information
Let agents work across your tools
Automate complex workflows
Keep your data and models under your control
Your existing tools
Your infrastructure
Company knowledgeAccess & policies
Agent executionCustom models
Self-hosted · Every action audited
Optional cloud models
Only masked context leaves
01 / 03
Train a specialist that beats the frontier
| Model | Quality (% of max achievable score) | Cost per 1,000 listings (USD) |
|---|---|---|
| Action Labs · Qwen3.5-9B + GRPO | 87.3 | 0.50 |
| Qwen3.5-9B base (untrained) | 64.2 | 0.50 |
| Claude Fable 5 | 75.9 | 111 |
| Gemini 3.1 Pro | 75.9 | 19 |
| GPT-5.6-sol | 71.4 | 29 |
| GPT-5.5 | 69.7 | 34 |
| GPT-5.5-pro | 70.3 | 172 |
Our trained 9B model reached 87.3% of the achievable score at $0.50 per 1,000 listings. Chart shows zero-shot results across 200 held-out episodes.
Fermisense,
Explore AI’s Navier–Stokes result
OpenAI reports an AI-generated proof of finite-time singularities under smooth forcing, with a formalization in Lean.
OpenAI,
Generate targeted evaluations with Bloom
Anthropic’s open-source framework generates scenarios to measure the frequency and severity of researcher-specified model behaviors.
Anthropic,
Work with researchers and operators
Fabian Hildesheim
Co-founder
AI research at Stanford HAI and enterprise agent work at McKinsey QuantumBlack
Joël Hainzl
Co-founder
Procurement agents at Tacto, AI investing at Fortino, and a16z Scout experience
Justinas Zaliaduonis
AI Research & Engineering
ICML-published research at Stanford, knowledge retrieval, and model training and evaluation
Wolfgang Nimführ
General Manager Europe
40+ years at IBM bringing AI, analytics and cloud into enterprise organisations
