In March 2025, the Association for the Advancement of Artificial Intelligence (AAAI), published a report on the Future of AI Research. The report, which was led by outgoing AAAI President Francesca Rossi covers 17 different AI topics and aims to clearly identify the trajectory of AI research in a structured way. As part of this [&#8230;]

في مارس 2025، أصدرت جمعية تعزيز الذكاء الاصطناعي (AAAI) تقريرًا شاملاً حول مستقبل أبحاث الذكاء الاصطناعي، بقيادة الرئيسة المنتهية ولايتها فرانسيسكا روسي. يتناول التقرير 17 موضوعًا رئيسيًا في الذكاء الاصطناعي، بهدف رسم مسار مستقبلي لأبحاث الذكاء الاصطناعي بطريقة منظمة.

En marzo de 2025, la Asociación para el Avance de la Inteligencia Artificial (AAAI) publicó un informe integral sobre el Futuro de la Investigación en IA, liderado por la presidenta saliente Francesca Rossi. El informe aborda 17 temas clave de IA, con el objetivo de delinear la trayectoria futura de la investigación en IA de manera estructurada.

En mars 2025, l'Association pour l'avancement de l'intelligence artificielle (AAAI) a publié un rapport complet sur l'avenir de la recherche en IA, dirigé par la présidente sortante Francesca Rossi. Le rapport aborde 17 sujets clés en IA, visant à tracer la trajectoire future de la recherche en IA de manière structurée.

In March 2025, the Association for the Advancement of Artificial Intelligence (AAAI) released a comprehensive report on the Future of AI Research, led by outgoing President Francesca Rossi. The report addresses 17 key AI topics, aiming to outline the future trajectory of AI research in a structured manner.

AAAI 2025 presidential panel on the future of AI research – video discussion on AGI

arXiv:2512.10449v1 Announce Type: cross 
Abstract: The landscape of scientific peer review is rapidly evolving with the integration of Large Language Models (LLMs). This shift is driven by two parallel trends: the widespread individual adoption of LLMs by reviewers to manage workload (the "Lazy Reviewer" hypothesis) and the formal institutional deployment of AI-powered assessment systems by conferences like AAAI and Stanford's Agents4Science. This study investigates the robustness of these "LLM-as-a-Judge" systems (both illicit and sanctioned) to adversarial PDF manipulation. Unlike general jailbreaks, we focus on a distinct incentive: flipping "Reject" decisions to "Accept," for which we develop a novel evaluation metric which we term as WAVS (Weighted Adversarial Vulnerability Score). We curated a dataset of 200 scientific papers and adapted 15 domain-specific attack strategies to this task, evaluating them across 13 Language Models, including GPT-5, Claude Haiku, and DeepSeek. Our results demonstrate that obfuscation strategies like "Maximum Mark Magyk" successfully manipulate scores, achieving alarming decision flip rates even in large-scale models. We will release our complete dataset and injection framework to facilitate more research on this topic.

أظهرت دراسة حديثة ضعف المراجعين العلميين المعتمدين على نماذج اللغة الكبيرة (LLM) أمام حقن الطلبات غير المباشرة، مع التركيز على إمكانية تغيير قرارات المراجعة من 'رفض' إلى 'قبول'. تقدم هذه الدراسة مقياسًا جديدًا يُعرف باسم Weighted Adversarial Vulnerability Score (WAVS)، وتقيّم 15 استراتيجية هجوم عبر 13 نموذج لغة، بما في ذلك GPT-5 وDeepSeek، باستخدام مجموعة بيانات تضم 200 ورقة علمية.

Un estudio reciente ha examinado la vulnerabilidad de los revisores científicos basados en Modelos de Lenguaje Grande (LLM) ante la inyección de prompts indirectos, centrándose en la posibilidad de alterar las decisiones de revisión de 'Rechazar' a 'Aceptar'. Esta investigación introduce una nueva métrica, el Weighted Adversarial Vulnerability Score (WAVS), y evalúa 15 estrategias de ataque en 13 LLM, incluyendo GPT-5 y DeepSeek, utilizando un conjunto de datos de 200 artículos científicos.

Une étude récente a examiné la vulnérabilité des examinateurs scientifiques basés sur des modèles de langage (LLM) face à l'injection de prompts indirects, en se concentrant sur la possibilité de modifier les décisions d'examen par les pairs de 'Rejeter' à 'Accepter'. Cette recherche introduit une nouvelle métrique, le Weighted Adversarial Vulnerability Score (WAVS), et évalue 15 stratégies d'attaque à travers 13 LLM, y compris GPT-5 et DeepSeek, en utilisant un ensemble de données de 200 articles scientifiques.

A recent study has examined the vulnerability of Large Language Model (LLM)-based scientific reviewers to indirect prompt injection, focusing on the potential to alter peer review decisions from 'Reject' to 'Accept'. This research introduces a new metric, the Weighted Adversarial Vulnerability Score (WAVS), and evaluates 15 attack strategies across 13 LLMs, including GPT-5 and DeepSeek, using a dataset of 200 scientific papers.

AAAI 2025 presidential panel on the future of AI research – video discussion on AGI

Was this article worth reading? Share it

Ready to build your own newsroom?