AI agents: when the supply chain must learn to decide without compromising its security
Agentic artificial intelligence redefines the boundaries of cyber risk at the heart of critical information systems. In the supply chain, the transition from decision support tools to autonomous systems capable of performing business actions introduces a systemic vulnerability: the spread of uncontrolled effects and the expansion of the attack surface. An analysis by Bart Van der Biest, Managing Director of SAP Benelux.
Artificial intelligence no longer only suggests choices or analyzes dashboards. Autonomous agents now take over at the heart of logistics chains to assess unforeseen events, adjust schedules, launch orders or redirect transport flows. This rapid evolution poses a new challenge for companies in terms of controlling the security of these systems.
For Bart Van der Biest, CEO of SAP Benelux, the real difficulty lies in the ecosystem that frames its actions. "Creating an agent is no longer so complicated. The complexity lies mainly in security, authorizations, governance, but also in platforms allowing agents to collaborate in a controlled manner. "
"Fragmented decision-making autonomy generates a risk of error amplification comparable to the "bullwhip" effect in flow management. An agent can circumvent compliance requirements or select uncertified third parties."
A radical break for security officials
The integration of AI agents capable of autonomously assessing operational contexts, executing orders or reconfiguring logistics flows is radically transforming threat modeling. "Until now, agents were confined to the role of analytical assistants, but today they have write and trigger rights within ERPs and transactional platforms," explains Bart Van der Biest.
The major danger therefore comes from decisions made in a closed vase. If an agent only seeks to resolve a stock shortage as soon as possible, he may be tempted to choose a cheaper supplier but without the required certifications. Even worse, when several automated agents react in a chain to each other's actions, a small anomaly at the beginning can cause massive disruptions at the end of the line, like the famous whiplash effect.
Operational Transparency
To avoid these excesses, companies will in the future have to orchestrate their tools end-to-end rather than stacking isolated solutions.
The security architecture must impose strict execution limits, arbitration mechanisms in the face of contradictory decisions and exhaustive logging.
This means anchoring the models in the reality of the trade thanks to clear rules, by giving them access to contracts, actual stocks and authorizations in force. Transparency then becomes the key word, as Marcel Bron, head of the supply chain at SAP Benelux, recalls: "We must be able to trace everything that is done. "
Autonomy therefore does not mean the absence of safeguards. "The objective is not to entrust the keys to the company to freewheeling algorithms, but to delegate specific operational tasks to them within a strictly monitored perimeter. Human teams retain the key role: setting limits, arbitrating unforeseen cases and ensuring overall consistency of decisions."
SAVE THE DATE - October 15, 2026 Hi Site, Grimbergen (Brussels)
