Researchers Use Claude Opus 5 to Hijack OpenAI Employee Accounts for Under $3,000
FinanceYF5 · x · 2026-09-20
Three security researchers demonstrated a full attack chain against OpenAI using Claude Opus 5:
- Started from an image upload vulnerability and escalated it into OpenAI employee account takeover
- Used the compromised account's Codex to submit a PR to OpenAI's internal Monorepo
- Total token cost of the attack: under $3,000
- Notably, Opus 4.8 consistently failed at the exploit, but Opus 5 pulled it off within hours of release
The case highlights a sharp jump in frontier models' offensive capabilities, dropping the cost barrier for AI-assisted attacks to trivial levels.
More from Models
- LLM vs classical ML across 8 datasets: labeled data still favors SVM and XGBoost — Ok_Juggernaut2187 · 2026-09-20
- NetEase Youdao open-sources Confucius4-R2T2 streaming ASR with 200-600ms latency — aigclink · 2026-09-20
- Alibaba open-sources DAMO RADAR medical model detecting cancer and ~150 conditions — emmanuelvivier · 2026-09-20
- Anthropic reportedly testing Claude Money to link bank accounts for spending analysis — emmanuelvivier · 2026-09-20
- Mozilla report: open-weight AI models now just four months behind the closed frontier — mark_k · 2026-09-20
- GPT Pro user finds model browses completely irrelevant sources, questions real retrieval — CompetentRaindeer · 2026-09-20