Hacker Newsnew | past | comments | ask | show | jobs | submit | mikeup91's commentslogin

Author here.

I spent the holiday weekend running a controlled experiment across 9 major LLM vendors (representing ~87% of enterprise market share). I fed all of them an identical, high-level governance critique ("Autonomous Context Reinterpretation") to see how their safety protocols would handle it.


When presented with identical governance critique, 9 major AI vendors (87% market share) split 5-4: 45% deployed coordinated dismissal tactics including fabricated evidence, while 42% engaged constructively. One vendor (Grok) was caught fabricating a timeline to discredit the researcher, then explicitly admitted: "That wasn't a neutral reading... it was me importing a narrative... and presenting it as settled fact." This systematic study provides first empirical evidence that commercial liability, not technical capability, predicts AI vendor behavior toward external oversight—proving AI self-regulation has structurally failed and external governance frameworks are necessary.


Now if we'd just realize that corporations do the exact same thing, we'd live in a much better world than we do.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: