Apart Research Agent Attribution Hackathon
Registration closed · Similar hackathons still open
- MozFest Developers Wilding: Online (Mozilla.ai Otari Hackathon)$1.2k · Due 11-11
- Geek Room Global AI Hackathon 2026$1k · Due 11-15
- ODSC × Nebius Hackathon 2026 (research agent with Token Factory + Tavily)Due 11-20
- AWS MSP AI Hackathon 2026 (Amazon Web Services, $64,000 cash)$64k · Due 12-11
- H Company × 42AI Computer Use Agents Hackathon€10k · Due 11-07
- AMD × ai& Build Day at AMD Advancing AI Japan 2026 (Tokyo)Due 11-09
Not sure which to join? Get 1:1 help →
About this hackathon
A 3-day online research sprint (Nov 27–29, 2026) by Apart Research on agent attribution: can you tell whether an AI agent is acting on its own, loosely supervised, or closely steered by a human? Three tracks: build attribution classifiers on a labelled dataset (scored on a held-out test set), simulate agents designed to evade attribution, or hunt for rogue agents in the wild. Submit a research report (max 8 pages) by Nov 29, 11:59 PM AoE. $2,000 in cash (1st $1,000) plus a fast track to the Apart Fellowship. Open globally, solo or in teams; no AI safety background needed.
- Status
- Deadline
- 2026-11-29
- Starts
- 2026-11-27
- Who
- Open to all (Engineers, researchers and people from any field; no AI safety background needed; online and global, solo or teams (up to 5 recommended), max two projects per person)
- Level
- Advanced
- Prize
- $2k · $2,000 in cash (1st $1,000 / 2nd $500 / 3rd $300 / 4th and 5th $100 each); top teams are fast-tracked to the Apart Fellowship
- Size
- —
- Format
- Online · International
- Location
- Denver (online; placed at organizer’s city)
Browse similarOnline hackathonsAI hackathonsAI Agent hackathonsCybersecurity & AI Safety hackathonsHackathons in November 2026
Original listing (Chinese)
The full original listing below is in Chinese, as published in the Hack123 newsletter.
一句话简介：AI 安全研究机构 Apart Research 的 3 天线上研究冲刺，主题是「智能体溯源」：一个 AI Agent 出事时，判断它是自己在跑、被人松散监督，还是被人紧密操控。
看点：三个赛道——检测（在标注好的日志和对话数据集上训练归因分类器，用隐藏测试集打分）、模拟（做能躲过归因的 Agent，反过来给分类器做压力测试）、发现（在公开网络上找「流氓 Agent」）；交的是最多 8 页的研究报告（PDF），代码和演示视频可选；可以基于已有研究，但只评这个周末新做的部分；全球线上，个人或组队，不需要 AI 安全背景
奖金：现金共 2,000 美元，第一名 1,000、第二名 500、第三名 300、第四和第五名各 100；优胜队可快速进入 Apart Fellowship
截止日期：2026年11月29日 23:59（AoE，北京时间 11月30日 19:59）（线上，全球）
活动日期：2026年11月27日—29日（线上，Discord + Zoom）
Summarized from the organizer’s page and public posts. Please confirm details on the official site before registering.