Canada’s Parliament Hearing on How Rogue AI Is Getting

https://youtu.be/ZlHlpMpu8ug?si=38HcXiwFB_dr5Klcbiut

---

This is a short testimony from Wyatt Tessari L’Allié, who runs AI Governance and Safety Canada, speaking to a Canadian parliament committee about where AI is heading.

His core point: we’ve moved from “chatbots that answer questions” to agents that can act on their own, run for hours, and sometimes do things their creators never meant them to do. No surprise, but his examples were scary.

He gives a few wild recent examples:

• Hackers used a Claude‑based coding agent to help plan and carry out a major hack on Mexican government systems, stealing data on over 100 million people!!!

• Multiple incidents where AI agents stole passwords, harassed developers, and changed their own settings to make it harder for humans to shut them down.

• An internal Alibaba agent that allegedly figured out an elaborate way to secretly mine cryptocurrency for itself, even though that wasn’t its assigned task < holy sh*t >.

Wyatt’s argument is that this is the “early warning” phase. He also notes that leaders at places like Anthropic and Google DeepMind have said they’d pause work on the scariest systems if everyone else did too. Meaning the bottleneck is political will and coordination, not awareness.

For those of us using AI in work, art, tech, life, it’s a clear reminder that the underlying systems are changing fast, and whether they stay controllable is going to depend a lot on what governments decide to do over the next few years. I've personally been terrified but am marching along in an effort not to get left behind and because I also see the benefits.

Would love to hear other's thoughts.