Liam Romanis on AI | YouCongress

Liam Romanis

Independent cybersecurity and AI security consultant; author of a draft model-card security standard.

This standard applies to any machine learning model published to a public model repository with a downloadable model card or README. [...] Safety evaluation [...] MANDATORY [...] Results of safety evaluations (e.g. TruthfulQA, BBQ, red-teaming outcomes).

AI Verified (Apr 2026)

Source:

Liam Romanis, “Proposed OWASP Model Card Security Standard,” Draft v0.1, April 2026.

20d ago

Quote authenticity verification history

Report this

AI Verified The cited GitHub proposal identifies Liam Romanis and contains the stored scope and mandatory safety-evaluation wording verbatim; month precision matches the April 2026 draft. · Hector Perez Arenas gpt-5 · 20d ago

replying to Liam Romanis

Other opinions from Liam Romanis

See all

Liam Romanis votes For

Require open source AI models to include safety benchmarks in release notes

This standard applies to any machine learning model published to a public model repository with a downloadable model card or README. [...] MCSS-1 (Baseline) | Any publicly published model | All HuggingFace models, open-source releases [...] Safety ev...

more AI Verified source (Apr 20, 2026)