EU AI Checker Highlights Compliance Gaps in Major AI Models

A new tool from LatticeFlow reveals that leading AI models struggle with EU regulations on cybersecurity and discrimination.

Developed by LatticeFlow, Compl-AI is an open-source framework evaluating AI models against the EU AI Act.
Models from Meta, OpenAI, and others scored well overall but showed vulnerabilities in specific areas like discriminatory output and cybersecurity.
OpenAI's GPT-3.5 Turbo scored low on discrimination, while Meta's Llama 2 struggled with cybersecurity resilience.
The EU AI Act, effective in stages over two years, imposes fines for non-compliance, pushing companies to align models with regulations.
The framework aims to guide AI developers in enhancing compliance and is open for contributions from the AI research community.