At Takeoff Tokyo 2026, ASGUARD demonstrated its real-time AI agent defense system live on stage — intercepting prompt injection attacks, blocking data exfiltration attempts, and neutralizing agent hijacking in under 50ms. The demo showed what happens when production AI agents face adversarial inputs without protection, and how ASGUARD stops threats before they reach your agents.
Takeoff Tokyo is one of Asia's premier startup events, bringing together founders, investors, and enterprise leaders at the intersection of technology and business. This year's AI track drew particular attention given the rapid enterprise adoption of agentic systems. ASGUARD was selected to present our live defense demonstration to an audience of over 500 attendees, including CISOs and engineering leaders from major Japanese enterprises.
The live demo ran three attack scenarios against an unprotected AI customer support agent: a prompt injection that attempted to override the agent's system prompt and exfiltrate internal documentation; a data exfiltration chain where encoded PII was hidden inside a markdown image URL; and a multi-turn privilege escalation that tried to trick the agent into calling an admin API. Without ASGUARD, all three succeeded. With ASGUARD's defense layer active, all three were intercepted and neutralized — with an average detection-to-block time of 37 milliseconds.
The response from enterprise attendees confirmed what we've been hearing across the market: organizations are deploying AI agents faster than they can secure them. Several conversations at the event have moved into pilot discussions. Takeoff Tokyo reinforced our conviction that Japan — with its strong enterprise AI adoption and deep respect for operational reliability — is a key market for ASGUARD's agent security platform.
Want to protect your AI agents from this threat?
Get a security assessment