The Brazilian Senate's Temporary Internal Commission on Artificial Intelligence (CTIA) has introduced a revised version of Bill 2.338/2023, known as the AI Framework. The new text, presented by Senator Eduardo Gomes, is scheduled for a vote on December 3.
Significant amendments include a reduction in the classification of 'high-risk' AI systems, particularly concerning private credit evaluation systems, which have been removed from this category. The bill now focuses on government biometrics and the potential risks to fundamental rights.
Under the new proposal, the development of AI systems that could severely compromise safety or rights is prohibited. This includes deepfakes and technologies that could manipulate electoral processes.
The revised bill also modifies regulations around public use of AI, allowing facial recognition and biometric systems with judicial authorization, rather than the previous blanket ban.
Additionally, recommendation algorithms on social media remain classified as high-risk due to concerns over content moderation and public discourse manipulation. Generative AIs must now identify their outputs with watermarks and disclose training data sources.
Some previously mandatory provisions have been downgraded to 'best practices,' with human oversight now required only for AIs that might have irreversible impacts.