Le pare-feu comme composant du plan de défense de l’IA

Xavier Bensemhoun, a security evangelist and strategic-accounts engineer for Check Point in eastern Canada, argues that the firewall already deployed in most networks, not a new appliance, should become the control point for securing AI traffic. He frames Check Point's AI Defense Plane as three layers (risk-surface visibility, governance, protection) applied across four control points, then narrows the talk to the firewall because it sits at the crossroads of all network traffic. He demonstrates two capabilities delivered through a firmware upgrade to R82.20: Workforce AI, which lets the firewall recognise which AI application and account type (personal versus corporate ChatGPT, DeepSeek) a request uses and apply allow, ask or block rules with data-loss-prevention checks, and AI Agent Security, which inspects traffic between an internal application and a public language model to catch and block malicious prompt injection before it reaches the model. He closes on MCP servers, described as a bridge between AI systems and infrastructure, with granular access rules for SSH, file-share and API calls made through the protocol. A Q&A covers minimum firmware versions, added latency, and fail-open versus fail-closed behaviour when the cloud diagnostic service is unreachable.