Preparing for the EU AI Act: A Forward-Looking Perspective on Video Surveillance from Sunell

As artificial intelligence continues to transform the video surveillance industry, regulatory frameworks are rapidly evolving to ensure safety, transparency, and respect for fundamental rights.

The European Union’s AI Act (EU 2024/1689), set to be fully enforced by 2026, represents one of the most comprehensive regulatory efforts globally. For manufacturers, integrators, and end users of AI-powered video systems, preparing for these requirements is becoming increasingly important.

This article provides a forward-looking overview of how the AI Act may impact video surveillance—and how Sunell is exploring technical approaches aligned with these emerging requirements.

Understanding the EU AI Act: What to Expect

The EU AI Act introduces a risk-based framework that categorizes AI systems into four levels:

• Unacceptable Risk (Prohibited)
  Certain uses of AI—such as indiscriminate real-time biometric identification in public spaces (with very limited exceptions), non-medical emotion recognition, and large-scale facial database scraping—are expected to be restricted or banned.

• High-Risk AI (Strict Requirements)
  Systems involving biometric identification, critical infrastructure, law enforcement, and public safety will likely face stringent compliance obligations.

• Limited Risk (Transparency Obligations)
  Applications such as people counting, traffic analysis, or behavioral detection may require user notification and explainability.

• Minimal Risk (Low Regulation)
  Basic video recording and transmission systems without AI decision-making functions are expected to remain largely unaffected.

For video surveillance providers, this framework signals a clear shift toward accountability, transparency, and privacy-first design.

Implications for Video Surveillance Systems

As regulatory expectations evolve, several key changes are anticipated across the industry:

What may be restricted or phased out

• Real-time, large-scale facial recognition in public environments

• Emotion recognition in non-medical scenarios

• Sensitive attribute classification (e.g., ethnicity, religion, political views)

• Unauthorized large-scale facial data collection

What will require stronger technical capabilities

• Privacy-by-design system architectures

• Explainable AI (XAI) and auditable decision logic

• Legally sourced, bias-mitigated training datasets

• Robust cybersecurity and secure device management

• Human oversight mechanisms in critical decision workflows

• Comprehensive logging and traceability

These trends highlight the need for re-engineering AI systems—not just upgrading features.

Sunell’s Technical Exploration for AI Act Readiness

In response to these emerging requirements, Sunell is actively exploring and developing technologies aligned with AI Act principles. While regulatory implementation is still evolving, several key directions are shaping our approach:

1. Privacy-First Architecture (Privacy by Design)

• Edge AI processing to reduce reliance on cloud transmission of sensitive data

• Dynamic anonymization, such as real-time facial masking when identification is not required

• Separation of biometric features and identity data

• Minimization of data collection based on specific use cases

These approaches aim to support data protection principles while maintaining system performance.

2. Explainable AI and Human Oversight

• Traceable alert generation and event reconstruction

• Configurable and auditable algorithm rules

• Role-based access control and multi-level authorization

• Human-in-the-loop mechanisms for critical decisions

This helps improve transparency and supports accountability in AI-assisted operations.

3. Data and Algorithm Governance

• Emphasis on legally sourced and well-documented training datasets

• Ongoing evaluation of algorithm fairness and bias mitigation

• Continuous monitoring and iterative model optimization

• Consideration of data subject rights aligned with GDPR principles

4. Security and System Resilience

• Secure boot and firmware integrity verification

• Encrypted data transmission (TLS) and storage

• Elimination of default credentials and enforcement of strong password policies

• Vulnerability management and secure OTA (over-the-air) updates

5. Lifecycle-Oriented Compliance Approach

Sunell is exploring a full lifecycle framework that includes:

• Risk assessment and system design

• Development, testing, and validation

• Deployment and operational monitoring

• Continuous updates and eventual decommissioning

This aligns with the broader direction of AI governance under the EU AI Act.

Potential Deployment Scenarios Under Future Compliance

Different application scenarios may require different approaches:

• Commercial and campus security
  Edge-based analytics combined with anonymization and transparent notification mechanisms

• Critical infrastructure (e.g., transportation, energy)
  Enhanced oversight, traceability, and alignment with high-risk AI requirements

• Regulated law enforcement use cases
  Strict adherence to legal authorization, limited scope, and controlled data retention

Why Early Preparation Matters

Although full enforcement of high-risk requirements is expected by August 2026, early preparation can bring several advantages:

• Smoother market access in the EU

• Increased trust from partners and customers

• Reduced redesign costs in later stages

• Alignment with global trends in AI governance

Looking Ahead

The EU AI Act represents a significant step toward responsible AI deployment. While implementation details will continue to evolve, the direction is clear: AI systems must be transparent, secure, and respectful of individual rights.

Sunell will continue to monitor regulatory developments and refine its technologies to align with emerging requirements—supporting customers in building future-ready, trustworthy video surveillance systems.