The Philippines' BPO and tech-enabled services sectors have been among the most aggressive adopters of AI agents in Southeast Asia. Customer service bots, invoice processors, data-entry automations, and workflow triggers have proliferated across call centers in Makati, Cebu, and Clark — often deployed by department heads or team leads who need a quick fix, not by the IT security team that would flag a credential sitting unattended for six months.
That is exactly the gap Arrakis Security is describing in global terms, but it maps neatly onto what Philippine enterprise IT managers have been quietly complaining about for at least two years. The pattern is familiar: a tool gets onboarded, a webhook gets created, an API key gets embedded, the project ships, the team moves on, and nobody revisits the permission layer until something breaks or a compliance audit arrives.
For Philippine businesses, the relevance is direct. The Data Privacy Act of 2012 and the National Privacy Commission's implementing rules already require that personal data be protected by "reasonable and appropriate" organizational and technical measures. A dormant AI agent holding a live credential that can be triggered by an unsigned external webhook is, in plain terms, a reasonable person would argue, neither reasonable nor appropriate. If that agent touches customer records — which in a BPO context it almost certainly does — the exposure is not theoretical. The NPC has demonstrated willingness to impose penalties, and a breach traced to a neglected automation would be a particularly unflattering finding.
What to watch next is how Philippine regulators and industry bodies respond as AI agent adoption matures beyond experimentation. The DICT has been drafting AI governance frameworks, and the BSP has issued circulars on technology risk management that, while written for banks, set expectations that spill over into fintech and tech-enabled services. The question is whether dormant-credential hygiene gets addressed explicitly or gets swept into broader "AI risk" language that is too vague to enforce.
For business owners and operations leads, the practical takeaway is simpler than any regulation. Audit your AI agent inventory the way you audit user access: who deployed it, what it can reach, when it last ran, and who can start it from outside your perimeter. The 24% figure in the Arrakis report — dormant agents startable by unsigned external webhooks — should be the number that keeps a CTO in Makati awake, not because it is shocking, but because it is plausible.