AI That Hunts Software Flaws Could Make You Safer Online — Eventually

When hackers break into your bank, phone, or even your car, they usually use a hidden software flaw. Finding these flaws has always been slow, manual work — and that has quietly given attackers an advantage. Think of a giant mansion with thousands of unlocked doors. Each side, hackers and security teams, is searching separate parts of the house. When attackers find a door nobody else knows about, they can enter without being detected.

Now AI is changing that. Older automated tools, called fuzzing, could only poke a small part of the software that runs hospitals, power grids, and phones. The new generation of AI, including a model called Mythos Preview, can scan much more of the code quickly and easily. The researchers estimate it found about 70 to 80 percent of the severe hidden flaws in the software they tested. That's huge — it's like finally giving the security team a complete map of every unlocked door.

The good news is this should eventually favor the people defending you. If AI finds nearly every flaw before attackers do, hackers run out of secret entry points. The bad news? There's a painful transition period in 2026 and 2027. When AI discovers a flaw, software companies need time to patch it. Hackers know this. So in the next couple of years, we may see more attacks based on flaws that everyone knows about — but nobody has fixed yet. It's like posting a warning sign about a broken lock but leaving the door untouched for months.

Long-term, the future looks safer. The researchers predict defenders will catch up, and new software will be built with fewer weaknesses from the start. But for ordinary people, the next few years may feel like a storm before the calm. Your best defense remains simple: install updates quickly and keep backups of important files.