Atlantic AI Institute (Dalhousie, A2I2)
Senate-approved Dalhousie research institute, relaunched as the Atlantic AI Institute on 2026-08-20, "an Atlantic hub for artificial intelligence and machine learning, headquartered in Dalhousie University". Its research index lists eleven areas including "Trustworthy, interpretable, and responsible AI" and "Cybersecurity, privacy, and safe AI", and its own bibliography carries a sustained frontier-LLM-safety line: "Representation Noising: A Defence Mechanism Against Harmful Finetuning" (NeurIPS 37), "Immunization against harmful fine-tuning attacks" (EMNLP Findings 2024), "Evaluating Defences against Unsafe Feedback in RLHF", "Limits of Convergence-Rate Control for Open-Weight Safety" (2025) and "Open Technical Problems in Open-Weight AI Model Risk Management", alongside machine-unlearning and model-editing evaluation work. 23 core members; the safety line runs through Domenic Rosati, Hassan Sajjad, Frank Rudzicz and Ga Wu.