# Writing & Talks

## Janus: Defending LLM Agents Against Indirect Prompt Injection at the Tool-Call Boundary

Talk · Black Hat Arsenal — SecTor 2026 · October 7, 2026 · Toronto, Canada

Co-presented Janus, an open-source security framework for autonomous LLM agents, with Evan Harris. The demonstration explored runtime tool-call interception and least-privilege policy enforcement as practical mechanisms for mitigating indirect prompt-injection risks.

Presented at Black Hat Arsenal, SecTor 2026, on October 7 in Toronto, Canada.

The presentation demonstrated runtime tool-call interception and least-privilege policy enforcement as practical mechanisms for mitigating indirect prompt-injection risks.

**Co-presenter:** Evan Harris.

- [Official listing](https://blackhat.com/sector/arsenal/schedule/index.html)

## From Words to Images: Is CoT Still Trustworthy?

Research writing · Whitebox Research · 2025

Co-authored research investigating reasoning faithfulness in multimodal language models. The research code and report document controlled comparisons of mathematically equivalent text and image-based problems, evaluating answer correctness and indicators of reasoning faithfulness.

I contributed to the experimental framework, evaluation pipeline, analysis, and research report. The full research description, including the models and evaluation setting, is available in the Research section.

- [Research code & report](https://github.com/whitebox-research/c2-proving-ground-martinez-cot)

## Detecting Security Posture Drift in AI-Driven Software Development via Continuous Threat Modeling

Related research talk · USENIX Security 2026 · Enigma Track · 2026

Related presentation on the broader continuous threat-modeling research, delivered by Preeti Ravindra. The talk explores how threat-model differencing can surface security posture drift as AI-assisted development changes application architectures, permissions, and configurations.

**Presenter: Preeti Ravindra.** The broader research was featured at USENIX Security 2026, Enigma Track. My involvement was in the collaborative research described in the Research section.

- [Presentation · Preeti Ravindra](https://www.usenix.org/conference/usenixsecurity26/presentation/ravindra)