Yixiong Hao

Alexander K. Saeri, Jess Graham, Michael Noetel et al.

Artificial intelligence poses many risks, ranging from familiar present-day harms to unprecedented and potentially catastrophic ones. Effective risk management requires prioritization: we must understand which risks are most severe, who is most vulnerable, and who is most responsible for addressing them. We report results from a three-round Delphi study conducted late 2025 with 272 international AI experts. Experts rated 24 AI risks on harm probability and severity, sector and actor vulnerability, actor responsibility, and overall concern. Experts estimated the five most severe harms in the next 5 years were likely to come from dangerous capabilities, competitive dynamics, weapons & cyberattacks (including CBRNE), power centralization, and false information. In a business-as-usual scenario, experts judged 18 of 24 risks as having a more than 10% probability of catastrophic outcomes (e.g., more than 1 million deaths or more than USD 100B in financial loss) in the next 5 years (2025-2030). In a scenario where pragmatic mitigations are implemented, experts still judged five risks as having a more than 10% probability of catastrophic outcomes: dangerous capabilities, weapons & cyberattacks, environmental harm, inequality & unemployment, and power centralization. All 24 risks were judged as being more than 5% likely to cause catastrophic outcomes. AI users and the general public were judged the most vulnerable to these risks, but experts assigned the highest responsibility for addressing them to general-purpose AI developers and governance actors (including governments, regulators, and standards bodies). Across most risks, experts identified information, finance, and national security as the most vulnerable sectors. These findings can guide AI risk prioritization and clarify expert expectations about who should bear responsibility for mitigation.

Stepan Shabalin, Ayush Panda, Dmitrii Kharlapenko et al.

Sparse autoencoders are a promising new approach for decomposing language model activations for interpretation and control. They have been applied successfully to vision transformer image encoders and to small-scale diffusion models. Inference-Time Decomposition of Activations (ITDA) is a recently proposed variant of dictionary learning that takes the dictionary to be a set of data points from the activation distribution and reconstructs them with gradient pursuit. We apply Sparse Autoencoders (SAEs) and ITDA to a large text-to-image diffusion model, Flux 1, and consider the interpretability of embeddings of both by introducing a visual automated interpretation pipeline. We find that SAEs accurately reconstruct residual stream embeddings and beat MLP neurons on interpretability. We are able to use SAE features to steer image generation through activation addition. We find that ITDA has comparable interpretability to SAEs.

Glenn Matlin, Parv Mahajan, Isaac Song et al. · gatech

Wargames are simulations of conflicts in which participants' decisions influence future events. While casual wargaming can be used for entertainment or socialization, serious wargaming is used by experts to explore strategic implications of decision-making and experiential learning. In this paper, we take the position that Artificial Intelligence (AI) systems, such as Language Models (LMs), are rapidly approaching human-expert capability for strategic planning -- and will one day surpass it. Military organizations have begun using LMs to provide insights into the consequences of real-world decisions during _open-ended wargames_ which use natural language to convey actions and outcomes. We argue the ability for AI systems to influence large-scale decisions motivates additional research into the safety, interpretability, and explainability of AI in open-ended wargames. To demonstrate, we conduct a scoping literature review with a curated selection of 100 unclassified studies on AI in wargames, and construct a novel ontology of open-endedness using the creativity afforded to players, adjudicators, and the novelty provided to observers. Drawing from this body of work, we distill a set of practical recommendations and critical safety considerations for deploying AI in open-ended wargames across common domains. We conclude by presenting the community with a set of high-impact open research challenges for future work.

Yixiong Hao, Ayush Panda, Stepan Shabalin et al.

Controlling the behavior of Large Language Models (LLMs) remains a significant challenge due to their inherent complexity and opacity. While techniques like fine-tuning can modify model behavior, they typically require extensive computational resources. Recent work has introduced a class of contrastive activation engineering (CAE) techniques as promising approaches for steering LLM outputs through targeted modifications to their internal representations. Applied at inference-time with zero cost, CAE has the potential to introduce a new paradigm of flexible, task-specific LLM behavior tuning. We analyze the performance of CAE in in-distribution, out-of-distribution settings, evaluate drawbacks, and begin to develop comprehensive guidelines for its effective deployment. We find that 1. CAE is only reliably effective when applied to in-distribution contexts. 2. Increasing the number of samples used to generate steering vectors has diminishing returns at around 80 samples. 3. Steering vectors are susceptible to adversarial inputs that reverses the behavior that is steered for. 4. Steering vectors harm the overall model perplexity. 5. Larger models are more resistant to steering-induced degradation.