EU AI Act compliance for high-risk systems delayed until December 2027

The 16-month extension for high-risk systems provides critical breathing room for streaming platforms integrating AI into sensitive business functions like recruitment and legal analysis. While the delay eases immediate pressure, the underlying requirements for automated logging and human oversight remain unchanged, necessitating early investment in governance infrastructure. For the broader streaming ecosystem, the December 2026 watermarking deadline will force a standardized approach to content authenticity, impacting how AI-generated marketing and deepfakes are distributed. Organizations must now prioritize building audit trails that retain input and output data for at least six months to meet future transparency mandates. Watch for the August 2025 rollout of general-purpose AI model obligations to see how foundational providers adjust their training data transparency.

The EU AI Act's enforcement infrastructure is taking shape even as deadlines shift. In February 2025, the European Commission published its first set of guidelines on prohibited AI practices under the Act, clarifying that social scoring and emotion recognition in workplaces fall under immediate bans that took effect in February 2025. For streaming companies deploying AI in content moderation, recommendation engines, or workforce management, these prohibited-practice guidelines establish hard boundaries that the Digital Omnibus delay does not relax. The Commission also confirmed that national market surveillance authorities must be designated by August 2025, meaning enforcement bodies will be operational well before the extended high-risk deadline.

On the business and compliance side, the European AI Office launched its general-purpose AI code of practice consultation in April 2025, inviting model providers including those serving streaming platforms to commit to transparency and copyright compliance standards ahead of the August 2025 GPAI obligations. Companies that supply AI models used in video generation, dubbing, or content tagging for streaming services will need to demonstrate training data provenance and respect copyright opt-outs. The code of practice is voluntary but carries significant reputational weight, and the Commission has signaled it will factor adherence into future enforcement decisions. Meanwhile, industry groups including DigitalEurope warned in March 2025 that the AI Act's conformity assessment requirements remain unclear for many sectors, a gap that the Digital Omnibus extension does not resolve and that affects streaming firms building proprietary AI tools classified as high-risk.

Technical standards for AI transparency in media are advancing in parallel. The Council of Europe adopted its Framework Convention on AI in May 2025, the first binding international treaty on AI, which includes provisions on transparency for AI-generated content that overlap with the EU's watermarking requirements. For streaming platforms distributing AI-generated or AI-modified content across European markets, the convergence of the EU AI Act's December 2026 watermarking deadline and the Council of Europe's broader transparency obligations creates a dual compliance track. The IEEE Standards Association also published updated guidance on AI content provenance in early 2025, recommending C2PA-style cryptographic manifests as a technical baseline for machine-readable detection, which aligns with the EU's requirement for AI-generated content labeling and gives streaming engineers a concrete implementation path before the grace period expires.