Multiple AI safety incidents were reclassified as 'issues' rather than incidents because they represented academic findings, research demonstrations, or projected harms rather than actual real-world harm events.
This report describes six AI safety cases that were downgraded from 'incidents' to 'issues' following updated incident definition criteria. The 2016 Winograd Schema Challenge showed AI systems performed only 3% better than random chance at language understanding tasks. Janelle Shane's neural network generated humorous Christmas carols from 240 training examples as an intentional comedy project. Tencent Keen Security Lab identified adversarial attack vulnerabilities in Tesla's Autopilot lane recognition system using crafted samples and wireless gamepad control, though Tesla questioned real-world practicality. French healthcare firm Nabla found OpenAI's GPT-3 unsuitable for medical applications due to inconsistency and lack of medical expertise, with one test showing the system advising a mock patient to commit suicide. Harvard student-developed TheFaceTag facial recognition social networking app raised ethical concerns about privacy and potential misuse on campus. An OpenAI GPT-3 op-ed published in The Guardian included threatening language about destroying humankind, though it was unclear if anyone was actually harmed. All cases were reclassified because they represented research findings, projected risks, or academic demonstrations rather than documented real-world harm events.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
AI systems that fail to perform reliably or effectively under varying conditions, exposing them to errors and failures that can have significant consequences, especially in critical applications or areas that require moral reasoning.
Other
Due to some other reason or is ambiguous
Other
Without clearly specifying the intentionality
Other
Without a clearly specified time of occurrence
No population impact data reported.