EU AI Act Enforcement Hits Milestone as South Africa Lags
Why It Matters
The divergence between the EU's strict prohibitory framework and the global South's sectoral approach will dictate how international AI companies navigate market entry and compliance across different legal jurisdictions.
Key Points
- The EU AI Act's first set of bans on unacceptable risk systems has been legally binding since February 2025.
- EU penalties for non-compliance are severe, reaching the higher of €35 million or 7% of global annual turnover.
- By August 2026, the EU will mandate transparency for chatbots and deepfakes alongside new requirements for high-risk sectors.
- South Africa is opting for a sectoral regulatory approach using existing bodies like the Competition Commission rather than a central AI authority.
- South Africa currently lacks legally binding bans on biometric surveillance, social scoring, or mandatory algorithmic audits.
As of March 2026, the European Union has completed its first full year of enforcing bans on 'unacceptable risk' AI systems under the landmark EU AI Act. These prohibitions, which became legally binding on February 2, 2025, target real-time biometric surveillance, social scoring, and manipulative AI practices. Violations currently carry penalties of up to €35 million or 7% of global annual turnover. While the EU prepares for the August 2026 deadline regarding transparency for high-risk systems, other nations like South Africa are pursuing a fragmented regulatory path. The South African Department of Communications and Digital Technologies is currently developing a National AI Policy expected for 2027/2028, opting for oversight through existing bodies rather than a centralized regulator. This highlights a growing global gap in mandatory algorithmic audits and transparency requirements.
The EU is officially cracking down on 'bad' AI, celebrating a year of banning things like social scoring and facial recognition in public. If a company breaks these rules, they face massive fines that could reach 7% of their total global income. Meanwhile, places like South Africa are taking a much slower, 'middle-of-the-road' approach. Instead of one big law, they are using existing rules and won't have a specific AI policy ready for a few more years. This creates a two-speed world where some countries have strict 'no-go' zones for AI while others are still figuring out the basics.
Sides
Critics
Argues that South Africa's lack of a centralized AI regulator and delayed timeline for binding bans is a mistake.
Defenders
Focused on active market checks and ensuring compliance with the first wave of prohibitions.
Neutral
Developing a middle-of-the-road National AI Policy that relies on existing laws and sector-specific guidelines.
Noise Level
Forecast
The EU will likely see its first major enforcement actions and fines by late 2026 as Market Surveillance Authorities conclude initial investigations. South Africa’s proposed 2027/2028 timeline for regulation is likely to be delayed due to the complexity of coordinating between multiple existing regulators.
Based on current signals. Events may develop differently.
Timeline
Projected SA Regulation
The earliest expected window for enforceable AI regulations to begin in South Africa.
High-Risk AI Deadline
Strict data governance and human oversight requirements become mandatory for high-risk EU AI systems.
Enforcement Status Report
The EU marks over a year of active bans while South Africa's draft policy remains in development.
EU AI Act Bans Take Effect
Prohibitions on unacceptable risk systems like social scoring and manipulative AI become legally binding.
Join the Discussion
Discuss this story
Community comments coming in a future update
Be the first to share your perspective. Subscribe to comment.