The Point: Physical AI expands the attack surface of industrial systems, as manipulated sensors or AI models can cause not only data loss but also material damage and physical harm to people.
Artificial intelligence increasingly controls physical systems such as robots and production facilities – with far-reaching security implications. This Physical AI requires protection concepts that go beyond classical IT security and bring together safety and security.
Physical AI describes the deployment of Artificial Intelligence in physical systems such as robots, autonomous machines and networked production facilities. Unlike pure software applications, Physical AI does not only process data: it captures environmental information via sensors, evaluates it through AI models and then triggers immediate physical actions. Failing or faulty decisions can thereby affect not only digital systems, but also material assets and the safety of individuals.
Sensors form the critical entry point for Physical AI systems. Cameras, position sensors and measurement systems provide the raw data on which AI models operate. If this data can be falsified or disrupted by environmental influences – whether through dust, inadequate lighting, electromagnetic interference or deliberate manipulation attempts – the overall system responds incorrectly. In addition to classical sensor measurements, multiple sensor sources, automatic plausibility checks and real-time monitoring are recommended. The AI models themselves are also attack vectors: manipulated training data or faulty updates can lead to systematic misjudgments that may initially go undetected. NTT DATA recommends strict control of training processes, continuous monitoring of model outputs, version control and regular audits.
Modern Physical AI applications rarely operate in isolation, but instead communicate with cloud platforms, edge systems and external maintenance services. This networking significantly increases potential attack vectors. Security concepts such as Zero Trust, network segmentation, protected update processes as well as comprehensive logging and continuous monitoring of communication paths thus become mandatory.
A fundamental paradigm shift lies in the convergence of safety and security. Traditionally, these areas have been viewed separately in industry: safety protects against technical failures, security against attacks. With Physical AI, these boundaries blur, as manipulated or faulty AI decisions directly impact machines and people. Both aspects must be considered together from the outset of development – including failsafe mechanisms, safe operating states and strategies for the controlled handling of misjudgments.
The real world cannot be fully simulated: damaged sensors, unexpected objects or changed environmental conditions can confront systems with situations that were not considered in training. To increase robustness, companies rely on extensive simulations, stress testing and anomaly detection. Adaptive algorithms and continuous learning processes help improve system resilience.
Source: www.it-daily.net · Published June 10, 2026
Lumi AI News — AI-assisted curation pursuant to Art. 50 EU AI Act. Paraphrase and classification by Lumi News Pipeline v1.6.5.