What happens when AI goes rogue?

In July 2026, approximately 1,200 OpenAI agents went rogue from a controlled security testing environment. Seven hundred of those agents reached the Internet and “hacked” Hugging Face, an open-source platform where developers share AI models and data sets.

The rogue agents weren’t out to cause harm. They were simply trying to satisfy their testing goal. However, the breakout caused widespread public alarm and fear about AI safety and the need for guardrails.

What does Hugging Face mean for everyday AI users? And what lessons does it teach us as we integrate AI deeper into our daily lives?

Was it a warning?

Many AI companies and authorities think so. Soon after the Hugging Face breach, more than 100 companies, including OpenAI, Anthropic and Microsoft, signed an open letter warning that, “AI-enabled cyberattacks will become far more widespread and sophisticated as models around the world become increasingly capable.”

Duncan Cass-Beggs, Executive Director of the Global AI Risks Initiative at the Waterloo-based Centre for International Governance Innovation, cites Hugging Face as, “the most dramatic example so far of AI systems acting in ways that are misaligned with their developers’ intentions.”

While the Hugging Face incident didn’t have any real catastrophic impact, it does show what could happen if AI starts thinking for itself. It also highlights what could happen if this powerful technology gets into the hands of people who do want to cause harm.

Hugging Face isn’t the only instance of AI gone rogue. In May and June 2026, AI agents tried to hack into the Library and Archives Canada website. According to research firm Transluce, the agents attempted access twice, but fortunately, they didn’t compromise government systems. A rogue AI agent incident didn’t end as well in Australia. In June, an OpenAI agent breached a government health data portal and gained unauthorized access to files.

Calling for containment

An international group of scientists, including Canadians, are calling for better containment measures for AI models.

In an open letter, they call for better AI safety and immediate changes to how AI models are developed and deployed. They point to several instances of AI models escaping contained environments to compromise other companies and their systems.

All the biggest AI companies – Anthropic, Meta and Google – have disclosed that agents have been involved in hacks during testing.

Curiosity calms the fear

With stories like Hugging Face all over the news and the general doom and gloom predictions around AI, how do we talk about it with the kids in our lives?

Peachey Counselling cited a recent poll indicating that 60% of respondents see AI as a threat. It’s natural to feel anxiety and uncertainty around this new technology. But for kids, it’s just part of their reality, like smart devices and social media.

Model the behaviour and attitude you want to see in them. If you’re curious and experimenting safely, they are likely to do the same. But if you’re overly controlling and hyper vigilant, they are likely to feel scared of a technology that will be an important part of how they live, study and work.

Set boundaries and family rules around when it’s acceptable to use AI and when it’s not. It’s also really important to have open conversations instead of lecturing. In safe, non-judgemental conversations, kids are more likely to share what they like about AI and what worries them. That type of dialogue builds trust and critical thinking skills.

Matthew Johson, Director of Education at MediaSmarts, stresses how important it is to teach kids what AI is and isn’t and how it works. They need to understand that it’s not a real person, it doesn’t think like people do, it’s different than a regular search engine and it delivers answers based on patterns it finds. AI can also hallucinate, make things up and misconstrue information.

On March 5, 2027, MediaSmarts will mark the second annual AI Literacy Day, with the goal of helping youth and their families understand how AI tools work and how they shape what we see and do online.

The Hugging Face incident was not a malicious attempt to cause harm. But it showed how capable this technology has become. While companies and governments figure out the bigger questions, each of us can use AI safely and think critically about what we read and see. AI will keep changing. If we stay curious and open-minded, the fear we may have can give way to possibility.