#11 — Singapore AI-for-Science, Omnimodal agent orchestration framework

What does a national strategy for AI in science look like? Singapore has just provided a new answer, and that's where we begin this episode of Research Agent. [LEAD] Singapore launched eight inaugural projects for the AI-for-Science Initiative. Singapore officially launched eight inaugural projects for its AI-for-Science Initiative (AI4S) on June 16, 2026. This national program is spearheaded by the National Research Foundation. The initiative aims to utilize artificial intelligence to facilitate scientific discovery. SOURCE MATERIAL: (AI for Science: NUS leads cutting-edge research with 4 major AI ..., 2026-06-18): NUS has secured four major projects under Singapore's S$120 million AI-for-Science Initiative (AI4S), reinforcing its position as a global leader in AI-driven scientific research. This achievement underscores the University's unique strengths in bridging advanced AI capabilities with world-class expertise across multiple scientific disciplines, such as advanced materials, computing, genomics and agriculture. On 16 June 2026, Singapore officially launched eight inaugural projects under AI4S, a landmark national initiative spearheaded by the National Research Foundation to harness the power of Artificial Intelligence (AI) to revolutionise scientific discovery. [LEAD (FOLLOW-UP)] A framework for omnimodal agent orchestration was presented. A new framework has been developed for omnimodal agent orchestration. This framework provides a structured approach for coordinating various types of AI agents across different modalities. SOURCE MATERIAL: (AI Intelligence Briefing — June 15, 2026 (corrected) • Buttondown, 2026-06-15): This AI Intelligence Briefing details several advancements and challenges in AI, focusing on agentic AI, higher education, and enterprise adoption. It covers Microsoft's Discovery platform for scientific research, the critical need for authentication in higher education due to agentic AI cheating, OpenAI's new partner network, and a framework for omnimodal agent orchestration. Additionally, it highlights a report indicating alignment between online students and faculty on generative AI use. [LEAD] Government Office for Science updated its AI 2030 scenarios The Government Office for Science (GO-Science) updated its AI 2030 scenarios. These scenarios, initially developed in 2023 and published in 2025, provide tools for policymakers to navigate the evolving AI landscape. The update incorporates recent advancements in AI capabilities, increased investment, and shifts in geopolitics since 2023. The scenarios are designed for exploring uncertainties, stress-testing, and developing policy, aiming to establish a shared foundation for cross-government thinking on the future of AI. SOURCE MATERIAL: (AI Scenarios 2030: Helping policymakers plan for the future of AI ..., 2026-06-15): The Government Office for Science (GO-Science) updated its AI 2030 scenarios, initially developed in 2023 and published in 2025, to help policymakers navigate the evolving landscape of AI. The updated scenarios account for significant advancements in AI capabilities, increased investment and adoption, and shifts in geopolitics since 2023. These scenarios are tools for exploring uncertainty, stress-testing, and developing policy, rather than predictions, aiming to provide a shared baseline for cross-government thinking on the future of AI by 2030. [LEAD] Hong Kong announced an artificial intelligence research and development institute will become operational. Hong Kong announced plans for an artificial intelligence research and development institute to become operational this year. This institute will focus on establishing AI standards. Its purpose also includes promoting the safe application of AI. SOURCE MATERIAL: (R&D institute 'to set the standards for safe AI use' - RTHK, 2026-06-16): Secretary for Justice Paul Lam stated that Hong Kong is committed to advancing digital developments through artificial intelligence while combating data misuse and privacy breaches, aligning with the nation's 15th Five-Year Plan. He announced that an artificial intelligence research and development institute would become operational this year to establish AI standards and promote safe application. Privacy Commissioner Ada Chung added that her office launched the International Data Privacy Academy to make Hong Kong an international hub for privacy professionals, emphasizing that AI development must be guided by safety and driven by application. [LEAD] Researchers found a scaling law for large language models' moral judgment. Researchers systematically evaluated 75 large language model configurations using the Moral Machine framework to measure alignment with human preferences in life-death dilemmas. They observed a consistent power-law relationship, where distance from human preferences decreased as model size increased. This relationship persisted even after controlling for model family and reasoning capabilities, indicating that moral judgment reliability emerges with computational scale. SOURCE MATERIAL: (Scaling laws for moral machine judgement in large language models | Royal Society Open Science | The Royal Society, 2026-06-17): Autonomous systems increasingly require moral judgement capabilities, yet whether these capabilities scale predictably with model size remains unexplored. We systematically evaluate 75 large language model (LLM) configurations (0.27–1000B parameters) using the moral machine framework, measuring alignment with human preferences in life–death dilemmas. We observe a consistent power-law relationship with distance from human preferences (D) decreasing as D ∝ S − 0.10 ± 0.01 (R2 = 0.50, p < 0.001) where S is the model size. Mixed-effects models confirm that this relationship persists after controlling for model family and reasoning capabilities. Extended reasoning models show significantly better [LEAD (FOLLOW-UP)] AI agents accelerate catalyst research. Artificial intelligence agents enhance catalyst research by significantly reducing the time for tasks such as density-functional theory calculations. This acceleration is achieved through cutting-edge machine learning models, large open-access datasets, and AI assistants. These combined technologies hold the potential to expedite the process from catalyst discovery to industrial deployment. SOURCE MATERIAL: (AI Agents Accelerate Catalyst Discovery, Bridging Computational ..., 2026-06-15): Artificial-intelligence agents are rewriting the playbook for catalyst research, turning tasks that once consumed days of density-functional theory (DFT) calculations into minutes. This leap comes from cutting-edge machine-learning models, massive open-access datasets, and AI assistants. The convergence of AI agents, expansive open datasets, and automated experimentation is already delivering tangible results, with the potential to significantly cut the time from catalyst discovery to industrial deployment. [LEAD (FOLLOW-UP)] The House GAAIA Discussion Draft proposes a federal AI governance framework The House GAAIA Discussion Draft introduces a comprehensive federal AI governance framework. This framework includes oversight for frontier models, standards for AI development, and provisions related to the AI workforce and federal regulatory review. It mandates risk management and incident reporting for advanced AI developers, with the aim of positioning the U.S. as a leader in international AI standards. The proposal also includes a three-year preemption of certain state AI development laws. SOURCE MATERIAL: (House GAAIA Discussion Draft Proposes Federal AI Governance Framework | ArentFox Schiff, 2026-06-16): The House GAAIA Discussion Draft proposes a federal AI governance framework, integrating frontier-model oversight, development standards, workforce provisions, and federal regulatory review. It outlines detailed obligations for advanced developers in risk management and incident reporting, while also aiming to establish the U.S. as a leader in international AI standards. The draft includes a three-year preemption of certain state AI development laws and may see changes before formal introduction. [LEAD (FOLLOW-UP)] Hierarchical agentic large language model reasoning drove scientific simulation and exploration efficiently. Researchers demonstrated a method using hierarchical agentic large language model reasoning for scientific exploration and simulation. This approach reduced the required atomistic simulations by up to 90% in chemical applications. The strategy produced more coherent and information-efficient search trajectories than single-agent, multi-agent, or stochastic baselines. SOURCE MATERIAL: (Hierarchical Multi-agent Large Language Model Reasoning for Autonomous Heterogeneous Catalyst Discovery, 2026-05-25): Artificial intelligence is reshaping scientific exploration, but most methods automate procedural tasks without engaging in scientific reasoning, limiting autonomy in discovery. We demonstrate that hierarchical agentic large language model reasoning can efficiently drive simulation and scientific exploration. Across two chemical applications, CO adsorption on Cu surface transition metal adatoms and on M–N–C catalysts, reasoning-guided exploration reduces required atomistic simulations by up to 90% relative to heuristic or random selection. Comparisons across single-agent, multi-agent, and stochastic baselines show that hierarchical strategies yield more coherent and information-efficient sea [LEAD] Researchers realized agentic reasoning strategies in the Materials Agents for Simulation and Theory in Electronic-structure Reasoning system. A multimodal system named Materials Agents for Simulation and Theory in Electronic-structure Reasoning (MASTER) was created. This system implements agentic reasoning strategies to translate natural language into density functional theory workflows. MASTER aims to accelerate heterogeneous catalyst discovery. SOURCE MATERIAL: (Hierarchical Multi-agent Large Language Model Reasoning for Autonomous Heterogeneous Catalyst Discovery, 2026-05-25): Artificial intelligence is reshaping scientific exploration, but most methods automate procedural tasks without engaging in scientific reasoning, limiting autonomy in discovery. We demonstrate that hierarchical agentic large language model reasoning can efficiently drive simulation and scientific exploration. Across two chemical applications, CO adsorption on Cu surface transition metal adatoms and on M–N–C catalysts, reasoning-guided exploration reduces required atomistic simulations by up to 90% relative to heuristic or random selection. Comparisons across single-agent, multi-agent, and stochastic baselines show that hierarchical strategies yield more coherent and information-efficient sea [LEAD] Researchers introduced Agentic CLEAR, an automatic evaluation framework for AI agents. Agentic CLEAR is an evaluation framework designed to assess AI agent behavior. It provides dynamic, granular textual insights into agent performance across system, trace, and node levels. The framework was tested across multiple benchmarks and agentic settings, demonstrating its ability to provide high-quality, data-driven feedback. Its results align with human-annotated errors and predict task success rates. SOURCE MATERIAL: (Agentic CLEAR: Automating Multi-Level Evaluation of LLM Agents, 2026-05-21): Agentic systems are becoming more capable: agents define strategies, take actions, and interact with different environments. This autonomy poses serious challenges for overseeing and assessing agent behavior. Most current tools are limited, focusing on observability with basic evaluation capabilities or imposing static, hand-crafted error taxonomies that cannot adapt to new domains. To address this gap, we present Agentic CLEAR, an automatic, dynamic, and easy-to-use evaluation framework. It produces textual insights into the agent behavior on three levels of granularity: system, trace, and node. Agentic CLEAR operates above the observability layer, enabling seamless integration and featurin [NOTABLE (FOLLOW-UP)] NUS secured four projects under Singapore's AI-for-Science Initiative. The National University of Singapore (NUS) secured four projects under Singapore's S$120 million AI-for-Science Initiative (AI4S). These projects combine advanced AI capabilities with specialized expertise across scientific fields. Examples include advanced materials, computing, genomics, and agriculture. SOURCE MATERIAL: (AI for Science: NUS leads cutting-edge research with 4 major AI ..., 2026-06-18): NUS has secured four major projects under Singapore's S$120 million AI-for-Science Initiative (AI4S), reinforcing its position as a global leader in AI-driven scientific research. This achievement underscores the University's unique strengths in bridging advanced AI capabilities with world-class expertise across multiple scientific disciplines, such as advanced materials, computing, genomics and agriculture. On 16 June 2026, Singapore officially launched eight inaugural projects under AI4S, a landmark national initiative spearheaded by the National Research Foundation to harness the power of Artificial Intelligence (AI) to revolutionise scientific discovery. [NOTABLE] BAAI unveiled Physis-v0.1. China's Beijing Academy of Artificial Intelligence (BAAI) released Physis-v0.1, the first general world foundation model. This model integrates information to understand and predict real-world behavior, encompassing physical laws, spatial relationships, and common-sense knowledge. It aims to enable AI systems to anticipate events and make causal decisions. The development supports advancements in embodied AI and robotics. SOURCE MATERIAL: (China unveils AI 'world model' that understands physical world - CGTN, 2026-06-14): China's Beijing Academy of Artificial Intelligence (BAAI) has unveiled Physis-v0.1, the world's first general world foundation model, at the 8th Beijing Academy of AI Conference. Unlike large language models, world models like Physis-v0.1 are designed to understand and predict how the real world behaves, integrating information from various sources to comprehend physical laws, spatial relationships, and common-sense knowledge. This development aims to enable AI systems to anticipate events and make decisions based on cause and effect, addressing current limitations of AI in real-world environments and paving the way for advancements in embodied AI and robotics. [NOTABLE] OpenAI launched the OpenAI Partner Network OpenAI introduced a new program called the OpenAI Partner Network. This initiative is designed for global partners to create, distribute, and implement AI solutions using OpenAI's technology. The company committed $150 million to support this partner ecosystem. SOURCE MATERIAL: (Introducing the OpenAI Partner Network | OpenAI, 2026-06-14): The OpenAI Partner Network is a new program designed for global partners to build, sell, and deliver AI solutions using OpenAI technology. OpenAI is investing $150 million to support this ecosystem, aiming to help partners more rapidly bring the benefits of AI to a broader range of organizations. [NOTABLE] Gartner published the "10 Best Practices for Optimizing Generative and Agentic AI Costs" report. Gartner released a report titled "10 Best Practices for Optimizing Generative and Agentic AI Costs" in March 2026. This publication offers guidance to enterprises. Its focus is on managing expenses for generative and agentic AI systems as organizations move from pilots to production environments. SOURCE MATERIAL: (Gartner Publishes 10 Practices to Optimize GenAI Costs | Let's Data ..., 2026-06-14): The March 2026 Gartner report "10 Best Practices for Optimizing Generative and Agentic AI Costs" lays out enterprise guidance on controlling spend as organizations move from pilots to production, according to Gartner's publication and coverage. SiliconANGLE published a guest column on June 14, 2026 summarizing the ten practices, including model-cost tradeoffs, AI sandboxes, model cards, governance, and balancing upfront customization vs ongoing inference costs (SiliconANGLE). Airia's blog post says it was cited in Gartner's section on AI gateways and quotes Gartner that "Through 2028, at least 50% of GenAI projects will overrun their budgeted costs due to poor architectural choices and lack [NOTABLE] Gartner projected that 50% of Generative AI projects will exceed their budgets by 2028. Gartner's report includes a projection regarding generative AI project costs. The report predicts that at least half of these projects will experience budget overruns by 2028. Gartner attributes these overruns to inadequate architectural decisions and a lack of operational knowledge. SOURCE MATERIAL: (Gartner Publishes 10 Practices to Optimize GenAI Costs | Let's Data ..., 2026-06-14): The March 2026 Gartner report "10 Best Practices for Optimizing Generative and Agentic AI Costs" lays out enterprise guidance on controlling spend as organizations move from pilots to production, according to Gartner's publication and coverage. SiliconANGLE published a guest column on June 14, 2026 summarizing the ten practices, including model-cost tradeoffs, AI sandboxes, model cards, governance, and balancing upfront customization vs ongoing inference costs (SiliconANGLE). Airia's blog post says it was cited in Gartner's section on AI gateways and quotes Gartner that "Through 2028, at least 50% of GenAI projects will overrun their budgeted costs due to poor architectural choices and lack [NOTABLE] US administration barred foreigners from accessing Anthropic's top AI models, citing national security concerns. The US administration implemented restrictions preventing foreign nationals from accessing Anthropic's advanced AI models, Fable 5 and Mythos 5. This decision is attributed to national security concerns and export controls on advanced technology. The policy has implications for global research initiatives and foreign workers in the field. SOURCE MATERIAL: (US asks Anthropic to block global access to top AI models: Why it matters, 2026-06-14): The US administration has barred foreigners from accessing Anthropic's top AI models, Fable 5 and Mythos 5, citing national security concerns and export controls over advanced technology. This decision has reignited tensions between Anthropic and the Trump administration, especially after the company was previously blacklisted for refusing to allow its AI models to be used by the US military for surveillance and autonomous weapons systems. The order impacts global research and foreign workers, leading some experts to criticize the use of "foreign national" as a criterion for restricting access to AI models. [NOTABLE] Google detailed its AI safety strategy in a report. Google's 2026 report outlines its approach to integrating safety, governance, and trust into advanced AI systems. It details how responsibility is embedded across the AI lifecycle, from research to monitoring, to manage frontier risks and agentic systems. The report discusses responsible AI's benefits in areas like healthcare, climate resilience, and scientific discovery, while also addressing preparations for Artificial General Intelligence. SOURCE MATERIAL: (Governing AI at Scale: Google's 2026 Responsibility Plan | AI ..., 2026-06-16): Google's 2026 report details its strategy to integrate safety, governance, and trust into increasingly powerful AI systems. It outlines how the company is embedding responsibility across the AI lifecycle, from research to post-launch monitoring, to manage frontier risks and agentic systems. The report also highlights the tangible benefits of responsible AI in areas like healthcare, climate resilience, and scientific discovery, while preparing for the uncertainties of Artificial General Intelligence (AGI). That's our report for this week. We'll track these stories and more in our next episode. Thanks for listening to Research Agent.