arXiv subjects

Michael Simpson

Publications and source records attributed to Michael Simpson.

7 recordsLinked to original sources

arXiv2026

Position: Preventing AI-Generated CSAM Necessitates New Approaches to AI Safety

Modern artificial intelligence (AI) systems present profound new risks to child safety. AI is increasingly being misused to create AI-generated child sexual abuse material, facilitate child sexual exploitation, and reduce barriers to harm. In this paper, we argue that protecting children from AI-facilitated sexual abuse requires new approaches to AI safety. Existing safety techniques assume data accessibility, transparency, and evaluation practices that are incompatible with the ethical and legal constraints surrounding child sexual abuse material. We examine how these constraints create new technical challenges, such as limitations on dataset auditing, red teaming, and fine-tuning prevention. In turn, we outline *15 open problems* in online child sexual exploitation and abuse across the AI development lifecycle, from dataset curation and model design to deployment and long-term maintenance. We propose targeted recommendations for researchers, developers, and policymakers to bridge the gap between theoretical AI safety and the realities of child protection. Our work aims to reframe preventing AI-facilitated child sexual abuse as a central, safety-critical dimension for AI research, motivating work that translates responsible AI principles into concrete safeguards against the exploitation of children.

cs.CY↗

arXiv2026

Evaluation without Generation: Non-Generative Assessment of Harmful Model Specialization with Applications to CSAM

Auditing the fine-tunes of open-weight generative models for harmful specialization has become a new governance challenge for model hosting platforms. The standard toolkit, generative evaluation via curated prompts or red-teaming, does not scale to platform-level auditing and breaks down entirely for domains like CSAM where generation is legally constrained. This motivates the Evaluation without Generation problem: assessing model capabilities without producing outputs. We argue that in such settings, capability must be inferred from the model's state, either its parameters or internal representations, rather than its outputs. We introduce Gaussian probing, a method that characterizes how LoRA adaptors perturb a model's internal representations by measuring responses to Gaussian latent ensembles. Unlike raw-weight baselines, Gaussian probing reliably distinguishes benign from harmful specialization without sampling outputs. We demonstrate effectiveness in high-risk domains, including detecting models specialized for child sexual abuse material (CSAM), where output-based evaluation is legally and ethically constrained. Our results show that Gaussian probing provides a scalable non-generative alternative for evaluating high-risk generative systems and remains robust to weight rescaling, a representative adversarial manipulation.

cs.LG↗

arXiv2022

Misinformation Mitigation under Differential Propagation Rates and Temporal Penalties

We propose an information propagation model that captures important temporal aspects that have been well observed in the dynamics of fake news diffusion, in contrast with the diffusion of truth. The model accounts for differential propagation rates of truth and misinformation and for user reaction times. We study a time-sensitive variant of the \textit{misinformation mitigation} problem, where $k$ seeds are to be selected to activate a truth campaign so as to minimize the number of users that adopt misinformation propagating through a social network. We show that the resulting objective is non-submodular and employ a sandwiching technique by defining submodular upper and lower bounding functions, providing data-dependent guarantees. In order to enable the use of a reverse sampling framework, we introduce a weighted version of reverse reachability sets that captures the associated differential propagation rates and establish a key equivalence between weighted set coverage probabilities and mitigation with respect to the sandwiching functions. Further, we propose an offline reverse sampling framework that provides $(1 - 1/e - \epsilon)$-approximate solutions to our bounding functions and introduce an importance sampling technique to reduce the sample complexity of our solution. Finally, we show how our framework can provide an anytime solution to the problem. Experiments over five datasets show that our approach outperforms previous approaches and is robust to uncertainty in the model parameters.

cs.SI↗

arXiv2018

Reverse Prevention Sampling for Misinformation Mitigation in Social Networks

In this work, we consider misinformation propagating through a social network and study the problem of its prevention. In this problem, a "bad" campaign starts propagating from a set of seed nodes in the network and we use the notion of a limiting (or "good") campaign to counteract the effect of misinformation. The goal is to identify a set of $k$ users that need to be convinced to adopt the limiting campaign so as to minimize the number of people that adopt the "bad" campaign at the end of both propagation processes. This work presents \emph{RPS} (Reverse Prevention Sampling), an algorithm that provides a scalable solution to the misinformation mitigation problem. Our theoretical analysis shows that \emph{RPS} runs in $O((k + l)(n + m)(\frac{1}{1 - \gamma}) \log n / \epsilon^2 )$ expected time and returns a $(1 - 1/e - \epsilon)$-approximate solution with at least $1 - n^{-l}$ probability (where $\gamma$ is a typically small network parameter and $l$ is a confidence parameter). The time complexity of \emph{RPS} substantially improves upon the previously best-known algorithms that run in time $\Omega(m n k \cdot POLY(\epsilon^{-1}))$. We experimentally evaluate \emph{RPS} on large datasets and show that it outperforms the state-of-the-art solution by several orders of magnitude in terms of running time. This demonstrates that misinformation mitigation can be made practical while still offering strong theoretical guarantees.

cs.SI↗

arXiv2014

Clearing Contamination in Large Networks

In this work, we study the problem of clearing contamination spreading through a large network where we model the problem as a graph searching game. The problem can be summarized as constructing a search strategy that will leave the graph clear of any contamination at the end of the searching process in as few steps as possible. We show that this problem is NP-hard even on directed acyclic graphs and provide an efficient approximation algorithm. We experimentally observe the performance of our approximation algorithm in relation to the lower bound on several large online networks including Slashdot, Epinions and Twitter. The experiments reveal that in most cases our algorithm performs near optimally.

cs.SI↗

arXiv2010

Larc: a State Collapse Theory of Quantum Measurement

This proposes a new theory of Quantum measurement; a state reduction theory in which reduction is to the elements of the number operator basis of a system, triggered by the occurrence of annihilation or creation (or lowering or raising) operators in the time evolution of a system. It is from these operator types that the acronym `LARC' is derived. Reduction does not occur immediately after the trigger event; it occurs at some later time with probability $P_t$ per unit time, where $P_t$ is very small. Localisation of macroscopic objects occurs in the natural way: photons from an illumination field are reflected off a body and later absorbed by another body. Each possible absorption of a photon by a molecule in the second body generates annihilation and raising operators, which in turn trigger a probability per unit time $P_t$ of a state reduction into the number operator basis for the photon field and the number operator basis of the electron orbitals of the molecule. Since all photons in the illumination field have come from the location of the first body, wherever that is, a single reduction leads to a reduction of the position state of the first body relative to the second, with a total probability of $m\tau_L$, where $m$ is the number of photon absorption events. Unusually for a reduction theory, the larc theory is naturally relativistic.

quant-ph↗

arXiv1995

Localisation and Nonlocality in Compound Energy-Momentum Eigenstates

A thought experiment considering conservation of energy and momentum for a pair of free bodies together with their internal energy is used to show the existence of states that have localised position while being eigenstates of energy and momentum. These states are applicable to all varieties of physical bodies, including planets and stars in free motion in the universe. The states are compound entanglements of multiple free bodies in which the momenta of the bodies are anticorrelated so that they always sum to zero, while their total kinetic energy is anticorrelated with their internal energies, so the total is a constant, E. The bodies are relatively localised while the total state has well-defined energy and momentum. These states do not violate Heisenberg uncertainty because the total centre of mass is not localised, hence the states naturally describe whole universes rather than isolated systems within a universe. A further property of these states, resulting from the form of the entanglement, is that they display nonlocality in the full sense of signal transmission rather than the more restricted Bell sense.

quant-ph↗