Hi, thank you for keeping this table on LLMs security startups.
Please add the following startup:
Geodesia , website: www.geodesia.ai
Description:
Geodesia.ai — A European frontier AI lab building the trust, safety and governance layer that every enterprise LLM deployment will need by August 2026, when EU AI Act enforcement begins. Fines: up to 3% of global turnover.
Our first product, Geodesia G-1, is a non-invasive runtime that wraps any open-source LLM via vLLM — without modifying weights — and lifts a small open model to the safety bar of frontier closed systems. On Gemma 4 E2B (~2B parameters), G-1 reaches HaluEval AUROC 0.92 and internal answer-safety AUROC 0.96 — matching or beating models orders of magnitude larger and closed-source.
Built in: Constitutional Intelligence, a four-layer policy stack rooted in EU fundamental rights and customisable per industry, jurisdiction and ethics committee. EU AI Act compliance out of the box: FRIA, Annex IV, audit chain, kill-switch, watermarking, human oversight — all auto-generated.
Patent pending. Founding team: 10,000+ research citations across Bari, Oxford and Stanford; serial entrepreneurs with multiple exits. G-1 is live with design partners across banking, insurance and healthcare. Three more products — agentic trust mesh, frontier alignment bench, sovereign frontier model — already in the lab.
Hi, thank you for keeping this table on LLMs security startups.
Please add the following startup:
Geodesia , website: www.geodesia.ai
Description:
Geodesia.ai — A European frontier AI lab building the trust, safety and governance layer that every enterprise LLM deployment will need by August 2026, when EU AI Act enforcement begins. Fines: up to 3% of global turnover.
Our first product, Geodesia G-1, is a non-invasive runtime that wraps any open-source LLM via vLLM — without modifying weights — and lifts a small open model to the safety bar of frontier closed systems. On Gemma 4 E2B (~2B parameters), G-1 reaches HaluEval AUROC 0.92 and internal answer-safety AUROC 0.96 — matching or beating models orders of magnitude larger and closed-source.
Built in: Constitutional Intelligence, a four-layer policy stack rooted in EU fundamental rights and customisable per industry, jurisdiction and ethics committee. EU AI Act compliance out of the box: FRIA, Annex IV, audit chain, kill-switch, watermarking, human oversight — all auto-generated.
Patent pending. Founding team: 10,000+ research citations across Bari, Oxford and Stanford; serial entrepreneurs with multiple exits. G-1 is live with design partners across banking, insurance and healthcare. Three more products — agentic trust mesh, frontier alignment bench, sovereign frontier model — already in the lab.