Final Text of the AI Act is out: Our Initial Thoughts
William Fry
|
Summary
AI Act's final text out; prohibits AI systems distorting behaviour, exploiting vulnerabilities, inferring biometric categories, evaluating/classifying individuals unfairly, creating facial recognition databases untargetedly and assessing criminal risk based solely on profiling. High-risk AI systems need efficient risk management, documentation, registration, appropriate mitigation measures for feedback loops, and strong human oversight. Act applies to EU-based deployers, AI system outputs used in EU, and general-purpose AI models. Non-compliance warrants heavy administrative fines. Businesses should evaluate AI systems for compliance, document meticulously, implement disclosure measures, invest in human oversight, and communicate transparently regarding AI deployment in workplaces.
