On 1 May, Australia’s ASD ACSC joined CISA, the NSA and the rest of the Five Eyes partners to release joint guidance on the careful adoption of agentic AI services. The headline message: don’t let an AI agent touch sensitive data or critical systems without strict guardrails. Around the same time, hospitality industry coverage flagged that hotel and venue software is being quickly rebuilt around agents that can take real actions on a venue’s behalf — booking changes, refunds, rate adjustments, guest messaging.

For an Australian hospitality operator, this isn’t an academic conversation. An AI agent inside your PMS or booking flow isn’t just answering questions. It can adjust inventory, issue refunds, change rates, and touch guest data covered by the Privacy Act. The difference between an agent that re-accommodates a guest within policy and one that decides to refund the room because the email “sounded upset” is governance you set — not magic the vendor ships. And a poisoned input — say, a supplier email or a scraped review — becomes a real-money problem, not a hallucination problem.

The practical play is fairly boring, which is the point. Map every AI feature your PMS, POS and booking platforms have switched on, often by default. Set permissions to read-only where you can. Require human sign-off for refunds, rate changes, and any guest-facing communication. Treat anything the agent reads from the outside world (emails, web searches, supplier feeds) as untrusted input. And run quarterly reviews — vendors are shipping new agent capabilities monthly, and the defaults change without anyone telling you. The ACSC guidance is a strong starting checklist.

If you’d like help auditing what your hospitality stack is letting AI do without you noticing, our Hospitality IT Services team does this work for venues across the eastern seaboard.