New GPT-5 Model Struggles to Hold Against Prompt Attacks

Do You Know?

In 2013, hackers breached Target’s payment network—not by attacking Target directly, but by stealing network credentials from its HVAC contractor. This “backdoor through the air-conditioning” trick compromised 40 million credit cards, showing that in cybersecurity, the weakest link might be the company you don’t even think about.

 🚨 This Week in Cyber + AI: Headlines You Shouldn't Miss

• 🔓 ChatGPT red-team discover simple tricks can still bypass GPT-5 safeguards

• 🚗 Dealer portal glitch gave access to thousands of connected cars

• 🚌 A free Wi-Fi loophole could let attackers track city buses in real time

• 🤖 DARPA tools now teach AI to find and patch software bugs autonomously

• 🌐Cyber expert warns AI misuse could undermine global information trust

🚗 Flaw in Dealer Portal Let Hackers Unlock and Start Cars Remotely
A vulnerability in a major automaker’s dealership system gave attackers the power to locate, unlock, and even start vehicles armed with nothing more than a license plate number.
🔍Why it matters: This wasn’t just data theft; criminals could have driven away with your car while you slept.
📎 Source: TechCrunch – Dealer Portal Car Hack

🚌 Free Wi-Fi on City Buses Gave Hackers a Live Map of Every Ride
Security researchers found that the “free Wi-Fi” on public buses was leaking network data—letting attackers see bus locations in real time and potentially hijack control systems.
🔍Why it matters: A tech convenience meant to help passengers could instead give attackers a blueprint for disrupting city transit. Imagine gridlocked streets… on purpose.
📎 Source: SC Media – Bus Wi-Fi Hacking Risk

🤖 DARPA’s AI Bug Hunters Could Become the Hackers of Tomorrow
DARPA’s AI Cyber Challenge produced systems that can find and fix vulnerabilities without human help, but the same tech could scan and exploit flaws faster than defenders can patch them.
🔍Why it matters: Tools that patch in seconds can also break in seconds—depending on who’s holding the controls.
📎 Source: CyberScoop – DARPA AI Challenge Winners

🌐 The ‘Information Dirty Bomb’ That Could Shatter Public Trust
Former Australian cybersecurity chief Alastair MacGibbon warns that state-built AI models could flood the world with convincing lies, leaving people unable to tell fact from fiction.
🔍 Why it matters: Once trust in information collapses, democratic institutions, markets, and even public safety could crumble alongside it.
📎 Source: The Guardian – AI Disinformation Threat