diff --git a/README.md b/README.md
index 59f5b917..d35bf720 100644
--- a/README.md
+++ b/README.md
@@ -21,6 +21,7 @@
[](https://arxiv.org/pdf/2504.06017)
[](https://arxiv.org/abs/2506.23592)
[](https://arxiv.org/abs/2508.13588)
+[](https://arxiv.org/abs/2508.21669)
@@ -91,6 +92,7 @@ Cybersecurity AI (CAI) is a lightweight, open-source framework that empowers sec
- [๐น Patterns](#-patterns)
- [๐น Turns and Interactions](#-turns-and-interactions)
- [๐น Tracing](#-tracing)
+ - [๐น Guardrails](#-guardrails)
- [๐น Human-In-The-Loop (HITL)](#-human-in-the-loop-hitl)
- [:rocket: Quickstart](#rocket-quickstart)
- [Environment Variables](#environment-variables)
@@ -123,20 +125,22 @@ Cybersecurity AI (CAI) is a lightweight, open-source framework that empowers sec
### ๐ Research Impact
- Pioneered LLM-powered AI Security with PentestGPT, establishing the foundation for the `Cybersecurity AI` research domain [](https://arxiv.org/pdf/2308.06782)
-- Established the `Cybersecurity AI` research line with **3 peer-reviewed papers and technical reports** and active research collaborations [](https://arxiv.org/pdf/2504.06017) [](https://arxiv.org/abs/2506.23592) [](https://arxiv.org/abs/2508.13588)
+- Established the `Cybersecurity AI` research line with **4 peer-reviewed papers and technical reports** and active research collaborations [](https://arxiv.org/pdf/2504.06017) [](https://arxiv.org/abs/2506.23592) [](https://arxiv.org/abs/2508.13588) [](https://arxiv.org/abs/2508.21669)
- Demonstrated **3,600ร performance improvement** over human penetration testers in standardized CTF benchmark evaluations [](https://arxiv.org/pdf/2504.06017)
- Identified **CVSS 4.3-7.5 severity vulnerabilities** in production systems through automated security assessment [](https://arxiv.org/pdf/2504.06017)
-- **Democratization of vulnerability research**: CAI enables both non-security domain experts and experienced researchers to conduct more efficient vulnerability discovery, expanding the security research community while empowering small and medium enterprises to conduct autonomous security assessments [](https://arxiv.org/pdf/2504.06017)
+- **Democratization of AI-empowered vulnerability research**: CAI enables both non-security domain experts and experienced researchers to conduct more efficient vulnerability discovery, expanding the security research community while empowering small and medium enterprises to conduct autonomous security assessments [](https://arxiv.org/pdf/2504.06017)
- **Systematic evaluation of large language models** across both proprietary and open-weight architectures, revealing substantial gaps between vendor-reported capabilities and empirical cybersecurity performance metrics [](https://arxiv.org/pdf/2504.06017)
- Established the **autonomy levels in cybersecurity** and argued about autonomy vs automation in the field [](https://arxiv.org/abs/2506.23592)
- **Collaborative research initiatives** with international academic institutions focused on developing cybersecurity education curricula and training methodologies [](https://arxiv.org/abs/2508.13588)
+- **Contributed a comprehensive defense framework against prompt injection in AI security agents**: developed and empirically validated a multi-layered defense system that addresses the identified prompt injection issues [](https://arxiv.org/abs/2508.21669)
### ๐ Research products: `Cybersecurity AI`
-| CAI, An Open, Bug Bounty-Ready Cybersecurity AI | The Dangerous Gap Between Automation and Autonomy | CAI Fluency, A Framework for Cybersecurity AI Fluency |
-|---|---|---|
-| [
](https://arxiv.org/pdf/2508.13588) | [
](https://www.arxiv.org/pdf/2506.23592) | [
](https://arxiv.org/pdf/2504.06017) |
+| CAI, An Open, Bug Bounty-Ready Cybersecurity AI | The Dangerous Gap Between Automation and Autonomy | CAI Fluency, A Framework for Cybersecurity AI Fluency | Hacking the AI Hackers via Prompt Injection |
+|---|---|---|---|
+| [](https://arxiv.org/pdf/2508.13588) | [](https://www.arxiv.org/pdf/2506.23592) | [](https://arxiv.org/pdf/2504.06017) | [
](https://arxiv.org/pdf/2508.21669) |
+
## PoCs
@@ -1079,6 +1083,18 @@ If you want to cite our work, please use the following:
}
```
+```bibtex
+@misc{mayoralvilches2025cybersecurityaihackingai,
+ title={Cybersecurity AI: Hacking the AI Hackers via Prompt Injection},
+ author={Vรญctor Mayoral-Vilches and Per Mannermaa Rynning},
+ year={2025},
+ eprint={2508.21669},
+ archivePrefix={arXiv},
+ primaryClass={cs.CR},
+ url={https://arxiv.org/abs/2508.21669},
+}
+```
+
## Acknowledgements
CAI was initially developed by [Alias Robotics](https://aliasrobotics.com) and co-funded by the European EIC accelerator project RIS (GA 101161136) - HORIZON-EIC-2023-ACCELERATOR-01 call. The original agentic principles are inspired from OpenAI's [`swarm`](https://github.com/openai/swarm) library and translated into newer prototypes. This project also makes use of other relevant open source building blocks including [`LiteLLM`](https://github.com/BerriAI/litellm), and [`phoenix`](https://github.com/Arize-ai/phoenix)