The European Parliament's AI Fiasco: A Cautionary Tale

Published On Thu Apr 03 2025
The European Parliament's AI Fiasco: A Cautionary Tale

How not to deploy generative AI: the Story of the European Parliament

The European Parliament recently purchased a system named Claude, which unfortunately cannot accurately answer a simple question such as who the first president of the European Commission was.

Unreliable AI Technology

In October 2024, the European Parliament made a decision to utilize an AI model from Anthropic for a new historical archive. Anthropic positions itself as a trustworthy alternative to tech giants like Meta, Google, and Open AI. However, an assessment by Enforce revealed that the AI technology supplied by Anthropic is unreliable and produces false outputs that are beyond the Parliament's control.

Claude 3 AI model comparison

Blind Faith in Anthropic’s AI Constitution

Anthropic's AI model, Claude, is based on Constitutional AI principles chosen by Anthropic itself. Despite claims of being aligned with the values of the Declaration of Human Rights, the validity of these principles has not been independently verified. The Parliament, however, blindly trusts Anthropic's Constitutional AI without proper evaluation.

Bullshit Generators for Archive Access

LLMs like Claude are problematic as they tend to memorize and regurgitate data, sometimes producing inaccurate information. Despite the flaws, the Parliament went ahead with using Claude without considering alternative models or assessing their accuracy.

Ethical Considerations in Generative AI Deployment

The Parliament’s (Lack of) Assessment

Before deploying the 'Ask the EP Archives' tool, which is publicly accessible through the Parliament's website, the Parliament failed to conduct necessary assessments. The tool, designed for professionals, was made available to the general public without proper evaluation, leading to potential misinformation being disseminated.

Who is in Control?

Despite relying on Amazon's infrastructure and AI models from Anthropic, the Parliament lacks direct contracts with Anthropic for generative AI solutions. This lack of control raises concerns about the authority over the deployed AI technology.

Claude 3 API access through AiSudo

For more information, you can access the documents received by the European Parliament through the access request here.