Author: Peter Chiang
Chief Consumer Experience Officer

The recent debate about AI safety raises a question that matters well beyond the technology industry: what happens when AI starts acting on our behalf? An AI agent is software that can take steps to complete a task for us, such as managing bookings or preparing a payment. Reports of agents taking unauthorized actions deserve careful investigation. People are right to expect clear answers about what went wrong and how it can be prevented.

The challenges are real. Our own AI assistant can misunderstand what we want or go further than we intended. Criminals can use AI to make scams more convincing and reach more people. And the personal information that makes an assistant useful—our relationships, routines, and business dealings—needs protection because it can also help someone deceive us. These risks overlap: knowledge of a family’s travel plans or a business’s suppliers can make a fraudulent request more believable to us or to an assistant acting for us.

Meeting these challenges takes technical understanding, rigorous engineering, and evaluations that measure how well protections work. Blanket assurances and sweeping predictions give people little guidance about the particular risks they face. We need enough detail to make informed decisions, including an honest account of what remains uncertain.

Recently, speaking in Australia as we introduced TrendLife Kaleida, I shared why these questions are personal to me. Becoming a father made me think about how AI might nurture my son’s curiosity and help him develop his own judgment. I also see possibilities for elders: technology that supports their independence and helps them stay connected to the people and activities that give their lives meaning, on their own terms.

For someone running a small business, that promise might mean finishing the paperwork earlier or having more time for customers. But the customer information, money, and reputation involved belong to a livelihood built over years. The safeguards need to reflect those stakes.

That is why I support Jensen Huang and NVIDIA’s emphasis on treating agent safety as an engineering responsibility. Trend Micro CEO Eva Chen has expressed confidence in cybersecurity’s ability to help make AI agents secure and useful. I share that conviction. At TrendLife, our job is to bring that discipline to the tools people use every day.

An important principle in NVIDIA’s approach is that instructions alone cannot be the boundary of an agent’s authority. We should teach AI to understand our intentions and respect our limits. We also need the surrounding system to enforce those limits when the AI makes a mistake or encounters someone trying to mislead it. The same care should apply to information: an assistant should only get the personal or business details it needs for the task.

Consider a business owner using an assistant to handle incoming invoices. It could check an invoice against an order and prepare a payment for review. Changing a supplier’s bank details or sending money should require the separate checks the owner has agreed to. A convincing email claiming to be from that supplier cannot grant the assistant new permission. The owner should be able to see what the assistant is allowed to do, change those permissions, and stop it when something looks wrong.

This is also where protecting personal context becomes essential. Imagine an elder receiving an urgent request for financial help that mentions a grandchild’s name and a real family trip. Those details can make the story feel credible, but they do not establish who sent it. A useful protection would help them verify the request through a contact they already trust, while leaving the decision in their hands. Knowing us better should make an assistant more useful. It should not make us easier to deceive.

Alignment also matters in quieter, everyday moments. A teenager asking AI for help with an essay may need a way to develop an idea without having the thinking done for them. They also deserve to understand what stays private and what a parent can see. A parent needs enough insight to support learning, while respecting the teenager’s growing independence. Keeping AI aligned with people means taking these different needs seriously and giving them a way to correct it when it misses the point.

Good engineering has to show its results. How often does a scam warning miss a fraudulent message, and how often does it flag a legitimate one? Can an agent still send money after its permission has been removed? Does a learning tool help someone understand an idea well enough to explain it themselves? We need measured results from realistic situations, with clear limits on what those results prove. The consequences matter too: missing a dangerous request and interrupting a harmless task carry different costs. Protection should reflect the action involved; setting a reminder and transferring money deserve different checks.

Some risks remain difficult to quantify. The International AI Safety Report documents substantial gaps in our knowledge. We should use those gaps to guide further testing, tighter limits, or decisions to delay a capability until its safeguards are adequate. Confidence should grow with the evidence.

This is the standard I want us to carry across TrendLife: protecting personal information, helping people recognize deception, and making AI and everyday digital tools safer to use. TrendLife Kaleida expresses that thinking in family life. As small businesses become part of TrendLife’s expanding mission, the same responsibility extends to the people building and running them. Families and business owners should be able to benefit from serious security engineering without having to become security experts themselves.

Every perspective matters because the choices we make as builders affect how other people live. I remain hopeful about what AI can make possible for my son, elders in our communities, and the people we serve. Earning their confidence takes the rigor of an engineer, the curiosity of a researcher, and the care of someone who understands what is at stake. That is the work I want TrendLife to be known for: helping people thrive with AI while keeping what matters to them safe.