You're listening to Tech Beat, your daily briefing on the stories shaping our digital world.
A security researcher this week laid out how granting an AI agent a single, seemingly harmless permission quietly opened a door for anyone who knew to knock. The piece is a sharp reminder that with agentic AI, the attack surface isn't just your system — it's every assumption baked into your trust model.
That theme of misplaced trust carries directly into a TechCrunch investigation finding that some Supabase customers have left substantial amounts of user data fully exposed on the open web. The culprit, researchers say, is a combination of AI-generated and vibe-coded applications where developers shipped fast and configured slow — leaving real people's personal information sitting in plain sight.
And on a very different note, cryptocurrency exchange Bitget is reporting losses now climbing to three hundred eighty seven million dollars after an attacker forged internal transfer requests to drain both hot and warm wallets. The exchange's CEO says the method and the fingerprints point toward North Korea, adding this to a growing ledger of suspected state-sponsored crypto heists.
Those are the stories worth watching today. Keep surfing. Tech Beat out.
