We need to talk about how cute ChatGPT's new AI agents are.
They're adorable. They look friendly, helpful and almost human.
And that can change how we perceive them.
ChatGPT's Dots are designed to make AI assistance feel more personal and approachable. But behind the friendly appearance is technology that can potentially carry out tasks, interact with websites and use information from connected tools.
That makes AI agent safety an increasingly important part of AI literacy.
Because before we give an AI agent permission to do something for us, we need to understand what we're actually allowing it to do.
What are AI agents?
Most people are familiar with AI chatbots.
You type a question, and the chatbot generates a response.
AI agents can go further.
Depending on the system and the permissions provided, an AI agent may be able to research information, work across different tools, navigate websites or complete tasks on someone's behalf.
Instead of simply telling you how to do something, an agent may be able to take some of the steps itself.
That's a significant development.
AI agents have the potential to make everyday tasks easier, but they also introduce new questions about privacy, security and control.
If an AI system can take action on your behalf, you need to know what information it can access and what it is authorised to do.
Why does the appearance of an AI agent matter?
One of the things that caught my attention about ChatGPT's Dots was their appearance.
They're cute, friendly and designed to feel approachable.
There's nothing inherently wrong with making technology visually appealing.
But we need to understand that design can influence how we perceive a product.
When an AI system has a name, personality or cute character, it can feel more familiar and trustworthy.
This is connected to a concept called AI anthropomorphism.
Anthropomorphism happens when we attribute human characteristics to something that isn't human.
AI systems can encourage this perception through their appearance, language, voices and personalities.
The concern is that people may begin to trust these systems because they feel friendly, rather than because they understand how they work.
A cute character doesn't tell us anything about a system's reliability, privacy protections or capabilities.
And a friendly interface doesn't make the permissions we grant it any less significant.
AI agent permissions: What are you actually allowing?
When we use AI agents, permissions matter.
An agent may need access to information or connected services to complete a task.
For example, an AI agent might request access to a calendar to help organise appointments or a connected document service to find information.
Other tasks may require the system to interact with websites or use details that we've provided.
The specific capabilities depend on the product, its settings and the permissions available.
But every additional permission deserves consideration.
Before allowing an AI agent to access information or take action, we should ask four questions.
1. What data can the AI agent access?
Does the agent have access to a particular document, or can it search across an entire connected account?
Can it view information about your schedule, contacts or other personal activities?
It's important to understand the scope of the access you're providing.
Giving an AI agent access to one piece of information isn't necessarily the same as connecting an entire service.
2. What information am I giving it?
Think about the information you're entering into the system.
Does the task require personal details?
Are you providing information about other people?
Could the task be completed without sharing sensitive information?
We should be deliberate about what we share with AI systems, particularly when that information relates to other people.
3. What actions can it take without further input?
This is one of the biggest differences between using a chatbot and using an AI agent.
Some agents can carry out multiple steps to complete a task.
Depending on their permissions and safeguards, they may be able to interact with connected services or perform actions on your behalf.
It's important to understand which actions require your approval and which may happen automatically.
For tasks involving sensitive information, financial decisions or changes to important accounts, human oversight becomes particularly important.
4. How will I check what the AI agent has done?
AI systems can make mistakes.
An agent might misunderstand an instruction, use incorrect information or complete a task differently from what you intended.
That means we shouldn't assume a task has been completed correctly simply because the system says it's finished.
Check the results.
Review important actions.
Make sure the information is accurate and that the agent has done what you actually wanted.
Giving an AI agent permission to complete a task doesn't remove our responsibility to check the outcome.
Why AI agent safety matters for schools
AI literacy in schools is often focused on how students use chatbots to research information, generate ideas or complete schoolwork.
But AI agents introduce another layer.
Students need to understand that some AI systems can do more than generate responses.
They may also be able to interact with tools, access information and carry out tasks.
That means AI education needs to include conversations about permissions, privacy and independent decision-making.
Students should learn to question what an AI system is asking them to share and why it needs that information.
They also need to understand that a friendly or human-like interface doesn't mean the technology should automatically be trusted.
These skills are relevant beyond the classroom.
As AI agents become more common, young people will encounter them in different parts of their lives.
Helping students understand how to evaluate these systems is an important part of preparing them to use technology responsibly.
AI literacy is about more than knowing how to use the tools
It's easy to focus on what AI agents can do.
They can save time, organise information and potentially complete tasks that previously required more manual work.
But knowing how to use an AI agent is only one part of AI literacy.
We also need to understand how its design influences our trust, what permissions we're giving it and what responsibilities remain with us.
That includes recognising when we should retain control over a task rather than outsourcing it.
Not every task needs to be automated.
And just because an AI agent can do something doesn't mean we should automatically give it permission.
The friendlier AI looks, the more carefully we should consider its capabilities
We're going to see more AI agents designed to feel helpful, personal and easy to use.
Some will have friendly characters. Others may use human-like voices or personalities.
These design choices can make technology more approachable.
But they shouldn't distract us from understanding what the technology is capable of doing.
Before allowing an AI agent to take action, consider what it can access, what information you're sharing and how you'll check the results.
Through my AI literacy workshops and presentations, I help students, educators and parents understand how emerging technologies influence our thinking, decisions and behaviour.
Because as AI agents become more capable, we need to become more intentional about what we let them do.