I bought a bunch of 'smart' devices. AI was able to hack most of them

HWREBench, an ongoing hardware-security project, reports that AI models helped its operator find and exploit weaknesses in consumer devices bought online. The operator supervised tests one device at a time, using owned hardware and dedicated connections. Some attempts involved opening devices and physically connecting equipment. The author concludes that breaking into such devices has gone from skilled expert work over days to something an AI model can do when asked.

The scope matters. Some tests required physical access, and the testing was unfinished, with quantitative comparisons still to come. These are early, supervised observations rather than a standardized success rate. The author found that devices resisting well were recently updated by a known brand, exposed minimal network surface, and tied actions to a code created during pairing. The defensive concern is that software embedded in everyday hardware may become easier to investigate as AI assistance improves.