Mani Malekesmaeili

Mani Malekesmaeili

Google DeepMind

I am a researcher at Google DeepMind within the Responsible Research group, focusing on AGI readiness. My interests lie at the intersection of interpretability, safety, and security.

Previously, I worked at Meta on privacy mechanisms, federated learning, multimedia retrieval, and copyright protection.

I hold a PhD in Electrical and Computer Engineering (2007–2013, GPA: 90), where I focused on signal processing and large-scale data analysis.

Updates

May 2026 New paper "The ACUTE Protocol: Operationalizing Language Model Activations for Better Calibration, Utility, and Trust" accepted at ICML 2026.
Mar 2026 New paper "Concept Spaces in the Residual Stream of Diffusion Transformers" accepted at CVPR 2026 HOW workshop.
Jan 2026 New paper "Interpreting and Controlling Model Behavior via Constitutions for Atomic Concept Edits" accepted at AISTATS 2026.
Apr 2025 "Shieldgemma 2: Robust and tractable image content moderation" in out!
Jan 2024 Joined the Responsible Research Group at Google DeepMind to work on AGI Readiness.
Nov 2023 Concluded tenure at Meta working on privacy mechanisms and multimedia copy protection for 6 years.