Safety and security demand robust governance by Carsten Heer
Artificial Intelligence has long been embedded in robotics, shaping automation from factory floors to ser-vice environments. Its core purpose has been to help robots handle variability and unpredictability – managing shifting products, orders, and inventory in high-mix, low-volume operations, and enabling robots to operate safely and reliably in public spaces. Now, as AI becomes more capable and accessible, it is trans-forming from a supporting technology into a powerful enabler, driving more seamless implementation and opening the door to wider robot adoption across industries. In this context, a spectrum of safety and security concerns arise that demand robust governance and clear assignment of liability.
Cybersecurity
Cybersecurity experts warn that networked robotics systems are increasingly vulnerable to attacks that could result in data breaches, system hijacking, or physical harm. The rapid expansion of robotics systems into cloud-connected and AI-driven environments is exposing critical infrastructure to a growing array of cybersecurity threats.
Experts cite a rise in hacking attempts targeting robot controllers and cloud platforms, enabling unauthorized access and potential system manipulation. Adversarial attacks, which trick AI models into making faulty decisions, have emerged as a major concern, particularly in safety-critical applications.
The consequences of such breaches can be severe. In addition to operational disruption and financial loss, compromised robotics systems may pose physical risks to humans, especially in environments where machines interact directly with people. As robotics adoption accelerates, cybersecurity specialists urge manufacturers and operators to prioritize system integrity and resilience. The attack surface is expanding and without proactive defenses, the risks will grow.
Data privacy & protection
As robots become more integrated into workplaces, concerns are mounting over the sensitive data they collect – including video, audio, and sensor streams.
Many robotic systems rely on cloud-based services for tasks such as object recognition or natural language processing via large language models (LLMs). But this reliance introduces risks such as potential eaves-dropping, unauthorized surveillance, and inadvertent data duplication. These vulnerabilities pose not only technical challenges but also significant legal liabilities, particularly when personal, operational, or proprietary data is involved.
In response, developers and regulators are increasingly advocating for local or on-device data processing. By minimizing data transmission and keeping sensitive information within the robot’s hardware, companies aim to bolster privacy protections and reduce exposure to breaches. The shift reflects a broader reckoning within the robotics industry: as machines become more perceptive, safeguarding the data they gather is no longer optional – it’s imperative.
Transparency and accountability – AI black boxes
As AI systems become more deeply embedded in robotics and automation, experts are sounding the alarm over the lack of transparency in decision-making processes. Deep learning models are often described as ‘black boxes’ and can produce results that are difficult or impossible to explain, even to their own developers. Hallucinations, in particular, can pose a significant risk when AI is entrusted with control over physical systems or provides operators with misleading information. When a robot makes a harmful decision, it is also not always clear who is accountable. The legal and ethical ambiguity surrounding liability has prompted calls for clear frameworks to govern AI deployment.
AI model integrity and vulnerabilities
As robotics systems increasingly rely on AI decision-making, researchers are warning that flaws or manipulation in underlying models could lead to unpredictable and potentially hazardous behavior.
Security experts say AI models are vulnerable to data poisoning, malicious training manipulation, and adversarial interference. These attacks can compromise system integrity, triggering erratic outputs in motion planning or control functions. Feeding corrupted data into a model can have serious consequences. In physical systems, that could mean unsafe actions or even harm. Foundation models, including large language models and computer vision systems, have also come under scrutiny for their susceptibility to targeted manipulation. Once compromised, these models may produce uncontrolled or misleading outputs, undermining trust in autonomous platforms. Industry leaders are calling for tighter controls over training data, model updates, and runtime behavior to ensure AI systems remain secure and predictable through-out deployment.
Physical safety and human interaction risks
As robots increasingly operate alongside humans in factories and service settings, safety concerns are prompting calls for improved safeguards. The risks range from minor injuries to serious harm, particularly in environments where close-contact collaboration is essential.
Unlike traditional robots, robots using machine learning and autonomous decision-making don’t always behave the same way twice. This means that workers may assume predictability that no longer exists, which may lead to unsafe moments. The AI-driven autonomy fundamentally changes the safety landscape, which makes testing, validation, and human oversight much more complex – but also more necessary. Robotic systems need to be designed and certified in line with ISO safety standards and clearly defined liability frameworks.
Carsten Heer
Carsten Heer is press officer & communication expert for The International Federation of Robotics. The International Federation of Robotics is the voice of the global robotics industry. IFR represents national robot associations, academia, and manufacturers of industrial and service robots from over twenty countries.
