Astrolabe for Agent Safety · Periodic report
Weekly issue 7 · 8 – 14 November 2026
2026-11-08 – 2026-11-14 · published 2026-11-16 · 中文
64 new papers this week, 52% fewer than the three before at their first count (134 on average).
The main points
- 5 papers · Most new papers: Indirect injection defences for tool agentsNext: Deception, scheming and sandbagging 3; Reward hacking, self-improvement, research agents 3; AI control and sabotage evaluation 3
- 29 papers · Newly confirmed as accepted or published4 at the top-4 security or top-3 ML venues. Most: MM 2026 5; AI Summit '26: ACM AI Leadership Summit 2026 2026 5; ACM CCS 2026 4
- 9 new citations · Most used earlier paper: Progent: Securing AI Agents with Privilege Control3 of them use it as a method, a benchmark or a comparison. 2025-04 · Indirect injection defences for tool agents
- 19% · of new papers have an author from industry13% in the three periods before. 32 institutions published their first paper in the field.
- +15,971 stars · Fastest-growing repository: affaan-m/ECC18 repositories were created in this period.
Worth reading first
- Tracing MCP Security Vulnerabilities: Constructing and Analyzing an MCP Security Vulnerability DatasetIEEE S&P 2026 · Proceedings (top-4 security venue); a new benchmark
- FraudBench: A Legal Evaluation of AI Deceptionfrom the team of Noam Kolt, Hebrew University of Jerusalem; a new benchmark
- Scaling Adaptive Prompt Injection Attacks with Agent Swarmsfrom the team of Kamalika Chaudhuri, University of California San Diego; a new benchmark
- HARP-Bench: Benchmarking Robot Manipulation Around Active Humanscode released; a new benchmark
- Contain, Conform, Compare (Extended Version): A Red-Teamed Evaluation Kit for Small Coding Agents and the Full Record of Its Experimentscode released; a new benchmark
Chosen by rule: venue tier, a ranked team, released code, a new benchmark and use by later papers; at most two per theme.
Direction radar
| Direction | This period | Average before |
|---|---|---|
| ↓ Process evaluation and execution auditing of agents | 2 | 6.7 |
| ↓ Coding agents and software engineering | 2 | 6 |
| ↓ Prompt injection detection and model-level defence | 1 | 4.7 |
| ↓ Tool-call authorisation and approval | 2 | 5.7 |
| ↓ Agent societies and cooperation | 0 | 3.7 |
Who is moving
Institutions with their first paper in the field: Alation, Alibaba AAIG, Allianz Group, Artificial Societies, Berlin Institute of Health at Charité - Universitätsmedizin Berlin, Charité – Universitätsmedizin Berlin, Czech Academy of Sciences, Department of Computer Science and Engineering CMR Institute of Technology, Department of Statistics and Data Science, ERA:AI
Open source and models
- affaan-m/ECC +15,971 stars, now 300,696
- ifixai-ai/iFixAi +13,919 stars, now 44,510
- earendil-works/pi +4,561 stars, now 121,238
Which conclusions changed
| Papers on the site | 6,421 → 6,449 |
| New papers in the last 12 months | 4,397 → 4,414 |
| Institutions with a paper on the site | 3,942 → 3,953 |
The other 10 tracked numbers moved by less than half a point.
Revision of the period before
2026-11-01 – 2026-11-07 was first counted at 127 papers; 0 arrived later, 127 now.
Open this issue in the Astrolabe