The AI Act is entering its operational phase, edge AI architectures are reconfiguring processing pipelines, and post-quantum cryptography is moving out of laboratories to settle into network stacks. These three axes, taken together, are reshaping the technical and regulatory constraints that IT teams face as we head into the fall of 2026.
Post-Quantum Cryptography on 5G Networks: Migration Without SIM Card Replacement
The transition to quantum-resistant algorithms typically encounters a hardware obstacle: the replacement of secure elements embedded in terminals. Thales has demonstrated that a post-quantum migration of 5G networks can occur without replacing SIM or eSIM cards, relying instead on a software update of the authentication layers.
This approach changes the economic calculus for operators. Deploying new algorithms on an existing fleet of millions of cards previously cost as much as the hardware renewal itself. Removing this constraint accelerates the migration window and makes a timeline aligned with the regulatory deadlines set for 2030 feasible.
We observe that this demonstration is prompting other equipment manufacturers to revise their roadmaps. Pressure is mounting on secure chip foundries, whose model relied on regular replacement cycles. For those wishing to explore the tech section of La Règle du Je, the topic of quantum security is regularly monitored there.
AI Act and Transparency Obligations: What Applies as of August 2, 2026
The European regulation on AI (Regulation (EU) 2024/1689) came into effect on August 1, 2024, but its application is proceeding in waves. As of August 2, 2026, the transparency obligations of Article 50 are effective for all providers and deployers of AI systems, including general-purpose models.
In practical terms, all AI-generated content (text, image, audio, video) must now carry a machine-readable technical marking. Content produced before August 2026 benefits from a catch-up window until December 2026 to comply with the marking, but new deployments must be compliant immediately.
Sanctions and Investigative Powers of the European AI Office
The European Commission and the European AI Office have investigative and sanctioning powers. Fines can reach 15 million euros or 3% of global turnover for general-purpose model providers who fail to comply with transparency rules.
A structuring point: the “omnibus measures package” adopted in July 2026 postpones most obligations for high-risk systems to 2027-2028. This creates a temporal gap between the transparency layer (active) and the enhanced technical compliance layer (deferred), which legal and technical teams must integrate into their planning.
- Content generation systems (text, image, video) are subject to technical marking now, regardless of company size.
- Systems classified as “high risk” (credit scoring, automated recruitment, biometric surveillance) see their obligations postponed to 2027-2028 by the omnibus package.
- General-purpose models (GPAI) must provide detailed technical documentation and comply with transparency rules under penalty of fines proportional to turnover.
Edge AI and IT/OT Convergence: Unification of Industrial Platforms
AI processing at the network edge is no longer an experimental concept. We are witnessing a shift in the manufacturing industry where decision-makers are targeting the unification of IT, OT, and physical security platforms. The goal: to reduce decision latency to a few milliseconds and eliminate dependence on the cloud for critical processes.
This convergence is changing the topology of industrial networks. Edge gateways now incorporate compact inference models, trained on digital twins, and then deployed directly on the production line. The cloud training/local inference cycle is becoming the de facto standard.
Predictive Maintenance and AI-Driven Intralogistics
Industrial maintenance in Europe still suffers from significant digital fragmentation. Sensor data flows into heterogeneous systems, rarely interoperable. Edge AI enables preprocessing and normalizing streams directly at the point of collection, reducing network load and accelerating anomaly detection.
In intralogistics, AI is reshaping the entire chain, from demand forecasting to the last meter in the warehouse. Autonomous mobile robots (AMR) carry their own navigation and trajectory planning models, without server calls. This local autonomy reduces failure points and increases operational resilience.
Humanoid Robots: Four Technical Barriers Before Industrialization
Humanoid robotics is attracting massive investments, but four major hurdles still hinder the transition to industrial scale. The fine manipulation of deformable objects remains an open problem: current grippers lack sufficient haptic feedback to handle soft materials without damaging them.
The second barrier concerns energy autonomy. Lithium-ion batteries limit humanoids to a few hours of continuous operation in real-world conditions, far from marketing promises.
- Dexterous manipulation of non-rigid objects, which requires sub-millimeter resolution pressure sensors.
- Insufficient energy autonomy for complete industrial work cycles.
- Sim-to-real transfer: behaviors learned in simulation degrade in uncontrolled environments, requiring costly fine-tuning on-site.
- Safety certification for human-robot coexistence in shared spaces, where current standards were not designed for mobile bipedal machines.
These barriers explain why deployments remain confined to controlled environments. Real industrialization will depend as much on regulatory maturity as on technological progress.
