𝐖𝐡𝐞𝐧 𝐀𝐈 𝐂𝐚𝐧 𝐀𝐜𝐭, 𝐖𝐡𝐨 𝐂𝐨𝐧𝐭𝐫𝐨𝐥𝐬 𝐀𝐈? With AI Agents, systems can now read data, call tools and take actions on behalf of users. That creates new opportunities, while it also changes the security requirements. The question is no longer simply “𝐖𝐡𝐚𝐭 𝐜𝐚𝐧 𝐀𝐈 𝐝𝐨?” It is “𝐖𝐡𝐚𝐭 𝐬𝐡𝐨𝐮𝐥𝐝 𝐀𝐈 𝐛𝐞 𝐚𝐥𝐥𝐨𝐰𝐞𝐝 𝐭𝐨 𝐝𝐨?”. A practical AI security approach therefore needs to look beyond the model itself and consider the full chain of interaction: 𝐃𝐚𝐭𝐚 - What information can the Agent access? 𝐏𝐫𝐨𝐦𝐩𝐭 - What instructions can influence its behaviour? 𝐌𝐨𝐝𝐞𝐥 - How is the model governed and constrained? 𝐀𝐠𝐞𝐧𝐭 - What actions is it permitted to take? 𝐓𝐨𝐨𝐥 - Which systems can it access, and under what conditions? This is where 𝐀𝐈 𝐆𝐮𝐚𝐫𝐝𝐫𝐚𝐢𝐥𝐬 - rules, policies and technical controls that keep AI systems safe and within defined boundaries, become critical. The goal is not to slow AI down, but to enable AI to 𝐚𝐜𝐭 𝐰𝐢𝐭𝐡 𝐚𝐮𝐭𝐨𝐧𝐨𝐦𝐲 𝐰𝐡𝐢𝐥𝐞 𝐫𝐞𝐦𝐚𝐢𝐧𝐢𝐧𝐠 𝐰𝐢𝐭𝐡𝐢𝐧 𝐢𝐭𝐬 𝐢𝐧𝐭𝐞𝐧𝐝𝐞𝐝 𝐬𝐜𝐨𝐩𝐞, reducing the risk of inappropriate data access, misuse of connected tools or unintended actions. There is another side to this equation: AI Agents can also strengthen cybersecurity itself. In Malware analysis, for example, Agents can support tasks ranging from extracting Indicators of Compromise and analysing configurations to reverse engineering and deobfuscation. This can help security teams process complex threats faster while keeping human expertise at the centre of investigation and decision-making. At 𝐕𝐆𝐒, we support businesses in exploring and applying AI Agent solutions by drawing on VNPT’s strong capabilities in both 𝐀𝐈 𝐚𝐧𝐝 𝐂𝐲𝐛𝐞𝐫𝐬𝐞𝐜𝐮𝐫𝐢𝐭𝐲, backed by specialized centers and subsidiaries with deep expertise in these fields. Through our advisory role, VGS connects businesses with the right expertise and solutions across the VNPT ecosystem, helping them 𝐜𝐚𝐩𝐭𝐮𝐫𝐞 𝐭𝐡𝐞 𝐩𝐫𝐨𝐝𝐮𝐜𝐭𝐢𝐯𝐢𝐭𝐲 𝐠𝐚𝐢𝐧𝐬 𝐨𝐟 𝐀𝐈 𝐰𝐡𝐢𝐥𝐞 𝐞𝐧𝐬𝐮𝐫𝐢𝐧𝐠 𝐭𝐡𝐚𝐭 𝐢𝐭𝐬 𝐮𝐬𝐞 𝐫𝐞𝐦𝐚𝐢𝐧𝐬 𝐬𝐞𝐜𝐮𝐫𝐞, 𝐰𝐞𝐥𝐥-𝐠𝐨𝐯𝐞𝐫𝐧𝐞𝐝 𝐚𝐧𝐝 𝐰𝐢𝐭𝐡𝐢𝐧 𝐭𝐡𝐞 𝐫𝐢𝐠𝐡𝐭 𝐛𝐨𝐮𝐧𝐝𝐚𝐫𝐢𝐞𝐬. 𝐑𝐞𝐥𝐚𝐭𝐞𝐝 𝐒𝐨𝐮𝐫𝐜𝐞: https://lnkd.in/gdscAvzC 𝐕𝐍𝐏𝐓 𝐆𝐥𝐨𝐛𝐚𝐥 𝐒𝐨𝐥𝐮𝐭𝐢𝐨𝐧𝐬 Build to operate. Trusted to deliver.