Divyanshu here.
I make AI systems efficient, secure and private, and find out what breaks when they are.
about
I’m an AI Research Engineer at Anaconda, which acquired Enkrypt AI in August 2026. At Enkrypt I was the founding ML research engineer and its lead AI safety and security researcher. Before that I was a Research Associate at IISc Bangalore, and I’ve been doing research since 2020. More about me →
research
I work on making AI systems efficient, secure and private, and on what breaks when they are made so: how quantization and fine-tuning weaken a model’s safety, red-teaming models across text, image and audio, and guardrails, differentially private fine-tuning and encrypted inference.
I maintain a public LLM safety leaderboard of about 300 models and an agent incident registry. My research has appeared at NeurIPS, AAAI and ICML workshops.
selected papers
- Lost in Serialization: Invariance and Generalization of LLM Graph Reasoners (GCLR @ AAAI 2026; GFM @ ICML 2026)
- Black-Box Red Teaming of Agentic AI: A Taxonomy-Driven Framework for Automated Risk Discovery (LAMaS Workshop @ AAAI 2026)
- Beyond Text: Multimodal Jailbreaking of Vision-Language and Audio Models through Perceptually Simple Transformations (NeurIPS Reliable ML from Unreliable Data Workshop, 2025)
- Quantifying CBRN Risk in Frontier Models (NeurIPS Reliable ML from Unreliable Data Workshop, 2025)
- No Free Lunch with Guardrails (arXiv preprint, 2025)
- SAGE-RT: Synthetic Alignment data Generation for Safety Evaluation and Red Teaming (NeurIPS Red Teaming GenAI Workshop, 2024)
- Increased LLM Vulnerabilities from Fine-tuning and Quantization (arXiv preprint, 2024)
news
- [Aug 2026] Anaconda acquired Enkrypt AI. Continuing my AI safety and evaluation research at Anaconda as an AI Research Engineer.
- [June 2026] Completed the Technical AI Safety course by BlueDot Impact.
- [Jan 2026] Presenting papers at the AAAI 2026 workshops: LAMaS, AIGOV, and GCLR. Will be sharing recent work and engaging in discussions on multi-agent systems, AI governance, and graph reasoning.
- [Nov 2025] Three papers accepted to AAAI 2026 workshops: one in LaMAS (LLM-based Multi-Agent Systems), one in AI Governance (AIGOV), and one in Graphs and Complex Structures for Learning and Reasoning (GCLR).
- [Sep 2025] Three papers accepted to NeurIPS 2025 workshops: one at LLM Evaluation and two at Reliable ML from Unreliable Data.