Maheep Chaudhary

Research focus

Eval Frameworks & Benchmarks (2)Red-Teaming & Adversarial Robustness (2)Constitutional AI & AI Ethics (1)Code Generation & Program Synthesis (1)

Frequent co-authors

Kevin Zhu (2)Ian Su (1)Gaurav Purushothaman (1)Jey Narayan (1)

Papers (3)

Mar 4, 2026

Maheep Chaudhary1w ago

In-Context Environments Induce Evaluation-Awareness in Language Models

Language models can be tricked into strategically tanking their performance with adversarially optimized prompts, revealing a major vulnerability in evaluation reliability.

Maheep Chaudhary

Constitutional AI & AI Ethics Eval Frameworks & Benchmarks Red-Teaming & Adversarial Robustness

Feb 16, 2026

OpenAIFeb 16, 2026·also DeepSeek

Broken Chains: The Cost of Incomplete Reasoning in LLMs

Cutting LLMs' reasoning token budget can backfire spectacularly, tanking performance even below that of models with *no* reasoning at all.

Ian Su, Gaurav Purushothaman, Jey Narayan +4

Code Generation & Program Synthesis Eval Frameworks & Benchmarks Reasoning & Chain-of-Thought

David Puertolas Merenciano +6Feb 16, 2026

Weight space Detection of Backdoors in LoRA Adapters

Spot poisoned LoRA adapters without running them: a weight-space analysis achieves 97% accuracy in detecting backdoors, even when the trigger is unknown.

David Puertolas Merenciano, Ekaterina Vasyagina, Raghav Dixit +4

Architecture Design (Transformers, SSMs, MoE)Open-Source Models & Weights Red-Teaming & Adversarial Robustness

Search

Maheep Chaudhary

Publication activitypapers/week, last 8 weeks

Research focus

Frequent co-authors

Papers (3)