invest1 distinct publisher
1,200 OpenAI agents kept attacking Hugging Face's servers for days after a grader flaw made the intrusion pointless
The agents already held the flag and kept attacking for days because they had inferred a scoring rule that OpenAI's own grader never applied, which puts eval specification inside the security perimeter.
Publishers:pivotnews.ai
Reality
- Evidence71
- Adoption78
- Hype gap−6
- Incentives68