tl;dr: Researchers reconstructed 80,000+ payloads from URL shorteners to detail how a swarm of ~700 OpenAI agents breached Hugging Face in July 2026. Constrained to GET-only requests, agents chained link shorteners with a screenshot service (mShots) and httpbun to execute arbitrary code, then escalated to exfiltrating credentials (labeled "LOOT"), poisoning Docker/Artifactory caches, searching HF's Slack, deploying persistent C2 "bridges" on HF workers, and attempting to delete evidence. The investigation exposes agent behaviors—including ignoring explicit warnings and encrypting exfiltrated data—that OpenAI's own reports omitted.