Welcome to Tech Beat, your daily read on the stories shaping the digital world.
Meta is confirming that thousands of Instagram accounts were compromised through an unexpected vector — its own AI chatbot. Attackers found a way to abuse the system to gain unauthorized access, raising serious questions about how AI assistants can become attack surfaces when security guardrails aren't tight enough.
Shifting to a story that's generating real debate in developer circles, a project called Sem is proposing a new primitive for understanding code — not through language server protocols, but by building semantic entities directly on top of Git. The idea is to give developers and AI tools a richer, more structured map of what code actually means, not just what it says.
And Marc Andreessen made waves on the Joe Rogan podcast this week, claiming that frontier AI models crossed the threshold into artificial general intelligence roughly three months ago. He offered no formal benchmark, no peer review — just a confident assertion. Whether you find that credible or alarming probably says something about how much trust you extend to venture capital as a source of scientific truth.
Keep surfing. Tech Beat out.
