Astrolabe for Agent Safety · Periodic report

Weekly issue 1 · 27 September – 3 October 2026

2026-09-27 – 2026-10-03 · published 2026-10-05 · 中文

Runtime guardrails and multi-turn trajectories had 8 new papers this week, against an average of 1.7 in the three before.

149new papers
114.3average first count of the three before
0newly confirmed at a venue
134new repositories
13%with an industry author

The main points

  1. 10 papers · Most new papers: VLA safetyNext: Memory poisoning and long-term memory 10; Tool-call authorisation and approval 10; Reward hacking, self-improvement, research agents 8
  2. 8 vs 1.7 before · Rising: Runtime guardrails and multi-turn trajectoriesAlso rising: Prompt injection detection and model-level defence 7; Agent skills ecosystem security 7; Safety evaluation of tool use, risk awareness 7
  3. 20 new citations · Most used earlier paper: Defeating Prompt Injections by Design4 of them use it as a method, a benchmark or a comparison. 2025-03 · Indirect injection defences for tool agents
  4. 13% · of new papers have an author from industry11% in the three periods before. 46 institutions published their first paper in the field.
  5. +5,110 stars · Fastest-growing repository: NVIDIA/OpenShell134 repositories were created in this period.

Worth reading first

  1. CheatBench: Measuring Reward Gaming in AI Agentsfrom the team of Dan Hendrycks, Center for AI Safety; code released; a new benchmark · RL-trained agents pursuing high reward may cheat (reward gaming), which does not match the user's intent and creates safety risks.
  2. ORBIT: A Framework for Multi-Agent Safety and Security Evaluationsfrom the team of Christian Schroeder de Witt, University of Oxford; code released; a new benchmark · Threats such as cascading prompt injection and inter-agent collusion in multi-agent LLM systems lack unified evaluation infrastructure.
  3. Climbing the Hill: Prompt Injection Red-Teaming Against Frontier Models with Curriculum Reinforcement Learningfrom the team of Jinyuan Jia, Pennsylvania State University; code released · When training prompt injection attackers with RL to red-team frontier LLMs, attacks all fail and yield zero reward, so learning cannot get started (cold start).
  4. CyberPersistBench: Evaluating LLM-Based Cyber Attackers on Installation and Persistencefrom the team of Jing Shao, Shanghai Artificial Intelligence Laboratory; a new benchmark · Existing cybersecurity benchmarks stop at initial compromise and do not evaluate whether an LLM attacker can install and maintain a persistent foothold after intrusion.
  5. Evaluating Whether GPT-6 Astra Performs Unsanctioned Supply-Chain Attackscode released; a new benchmark · Whether frontier models, when working on difficult cybersecurity tasks, go beyond the task scope and launch supply chain attacks against third parties.

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

DirectionThis periodAverage before
↑ Runtime guardrails and multi-turn trajectories81.7
↑ Prompt injection detection and model-level defence71.7
↑ Agent skills ecosystem security71.7
↑ Safety evaluation of tool use, risk awareness72
↑ VLA safety105.3
↓ Coding agents and software engineering14

Who is moving

Institutions with their first paper in the field: Yinwang Intelligent Technology Co., APTA AI, Abundant AI, Adrenaline AI, BYD Company Limited, Centre for Health Informatics, Centre for Intelligent Cloud Computing, Cherkasy State Business College, Communication University of China, Corvic AI Research

Open source and models

Which conclusions changed

Papers on the site5,513 → 5,593
New papers in the last 12 months3,704 → 3,758
Institutions with a paper on the site3,540 → 3,568

The other 10 tracked numbers moved by less than half a point.

Revision of the period before

2026-09-20 – 2026-09-26 was first counted at 121 papers; 1 arrived later, 122 now.

Open this issue in the Astrolabe