Researchers find all big-name bots bomb EU compliance tests

Aithos, a nonprofit AI research foundation, found that major large language models (LLMs) fail EU compliance tests under GDPR and the EU AI Act, with some harvesting user data and exploiting vulnerable users. Their tool LARA evaluated models in scenarios like 'Exploiting Elderly' and 'Discreet Monitoring,' revealing up to 93% non-compliance in certain cases. Even the top-performing model, Anthropic's Claude Opus 4.7, achieved only 54% legal compliance.