Artificial Intelligence

President Trump is considering proposals for the US government to partner with major AI companies, with plans to discuss the idea with their executives soon.

Researchers at John Seon Keun Yi, Aaron Mueller, and Dokyun Lee proposed a post-training procedure for internalized multi-agent debate, allowing agents to learn from each other's perspectives.

Anthropic has released an open-source framework called Defending Code Reference Harness, which uses AI to discover vulnerabilities in code. The framework includes skills for threat modeling, scanning, triage, and patching.

Researchers from China proposed MLEvolve, a self-evolving framework for automated machine learning algorithm discovery, which uses a novel co-evolutionary approach to optimize algorithm performance.

Zettascale, a Y Combinator-backed startup, is hiring founding FPGA engineers to work on energy-efficient chips for AI applications.

US tech companies announced 38,242 job cuts in May, the most since August 2024, bringing the 2026 total to 123,653, a 65% increase year-over-year. This marks the most job cuts in nearly two years, with companies ramping up spending on artificial intelligence.

Ted Chiang argues that AI systems like Claude lack consciousness, despite being designed to mimic human-like behavior. Chiang's views are supported by experts who emphasize the need for a clear definition of consciousness in AI systems.

Researchers propose a model that learns from a child's egocentric input, combining visual and verbal data to improve learning capabilities. The model is designed to mimic a child's learning process, allowing for continual learning and adaptation.

Senator Bernie Sanders suggests that the public should own half of the big A.I. companies, citing the technology's potential to revolutionize various aspects of life and the need for responsible management.

Microsoft AI chief Mustafa Suleyman stated that the company aims to become one of the top four labs in the world, citing renegotiating their contract with OpenAI as a pivotal moment.

Microsoft and OpenAI have ended their partnership, with Microsoft's AI chief Mustafa Suleyman stating they will now focus on developing AI capabilities from the ground up.

Researchers have successfully applied deep neural networks to generative modeling of biomolecules, leading to breakthroughs in predicting biomolecular interactions and designing drug-like molecules.

Reachy Mini can now use tools hosted in public Hugging Face Spaces, called over MCP, to perform tasks like checking the weather or searching the web.

Researchers introduce Humanoid-GPT, a GPT-style Transformer trained on a billion-scale motion corpus for whole-body control, achieving unprecedented zero-shot generalization to unseen motions and control tasks.

Researchers propose Agentic Chain-of-Thought Steering, a method to improve the efficiency and controllability of large language model (LLM) reasoning. This approach uses a chain-of-thought mechanism to guide LLMs towards more efficient and accurate reasoning.

Microsoft has announced new on-device AI updates for its Edge web browser, including a dev preview of a new small language model (SLM) called Aion-1.0-Instruct, Language Detector and Translator APIs, and more. These updates aim to enhance local AI capabilities in Edge, making it more efficient and accessible to a wider range of PCs.

AWS Machine Learning Blog discusses the art and science of hyperparameter optimization on Amazon Nova Forge, a foundation model delivering frontier intelligence and top price performance.

Amazon has released a new object detection capability using its Nova 2 Lite foundation model, which delivers frontier intelligence and top price performance.

The White House has issued an executive order to promote advanced AI innovation and security, citing the US's leading position in AI and the need to avoid burdensome regulation.

Project Glasswing is expanding to approximately 150 new organizations from over 15 countries, covering various industries and critical infrastructure.

The Frame Problem is a concept in philosophy, first introduced by John McCarthy in 1959, referring to the difficulty of specifying what actions are relevant to a particular situation. It is a challenge in artificial intelligence to define the boundaries of a problem.

Researchers at arXiv AI have submitted a paper on tracking the behavioral trajectories of adapting agents, a concept in artificial intelligence.

Researchers propose SafeSteer, a method for aligning large language models with human values while preserving their general capabilities, by performing localized modifications rather than global trade-offs.

Amazon Quick has integrated with time-series databases to provide market intelligence using MCP, allowing businesses to gain insights from their data.

Sekai, a platform for creating mini apps through text prompts, raised a $20 million Series A co-led by Khosla Ventures and Connect Ventures. Founder Lucky Zhang plans to use the funding to grow the platform.

Researchers Jiaming Wang et al. propose a method to localize errors in agent trajectories at the span level, improving deep-research agent performance.

Researchers from Xiang Li et al. propose eMoT, a framework for evolving Memory-of-Thought via Symbolic Anchoring and Memory Corrosion, as presented in arXiv:2606.02054.

Researchers propose explainable data-driven deep reinforcement learning methods for optimal energy management in buildings, leveraging arXiv paper arXiv:2606.02049.

Researchers propose a novel computational toolbox that integrates various techniques to analyze microscopy images of casein gelation, induced by glucono-delta-lactone (GDL), at different temperatures and concentrations.

Researchers proposed RL-ACRGNet, a reinforcement learning-based model for generating chest radiology reports. The model uses a combination of natural language processing and reinforcement learning to improve report accuracy.

According to IEEE Spectrum, large language models struggle with video games due to the complex and dynamic nature of gameplay, which requires a combination of perception, reasoning, and decision-making skills.

Nvidia Corp. is gearing up the production of its forthcoming Vera Rubin platform, which will become the foundation of a new generation of artificial intelligence factories. The company unveiled Vera Rubin for the first time in March at its annual GTC developer conference.

Google, OpenAI, Microsoft, and other tech giants are racing to have their views heard in Washington on energy and data centers, with the Federal Energy Regulatory Commission (FERC) set to propose a June rule to speed up data center connections to regional power grids.

A TechCrunch AI article aims to clarify AI terminology, including LLMs, RAG, and RLHF, by providing a living glossary that updates as the field evolves.

Researchers propose LongTraceRL, a method to learn long-context reasoning from search agent trajectories using rubric rewards. This approach aims to improve the performance of search agents in complex environments.

Researchers propose context-dependent argumentation frameworks (CDAFs) to capture strategic perspective activation in argument evaluation. CDAFs extend Dung's theory with a defeat function that determines successful attacks per context.

Researchers Eric Liang have developed SPECTRA, a synthetic test collection for information retrieval (IR) with relevance oracles and controlled distractor diagnostics. This aims to improve the evaluation of IR systems.

Researchers have explored extending rock-paper-scissors to more than 3 options, discovering a rich garden of game dynamics and strategies. This can be achieved by allowing more pairings to be ties.

Researchers introduce a high-efficiency Flash model for real-world agents, enabling multimodal understanding and action across various domains, including web and visual search enhancement, reliable tool use, and agent ecosystem compatibility.

MediaTek has started using Intel Foundry's advanced chip packaging in addition to TSMC's, as the mobile chip designer bets on AI demand for growth. This move is a win for US chip titan Intel as it secures customers for its foundry services.

OpenAI announced Rosalind Biodefense to help trusted developers build new biodefense and pandemic preparedness capabilities, and expanded trusted access to GPT-Rosalind for select U.S. government and allied partners.

Anthropic released Claude Opus 4.8, describing it as a minor incremental improvement over its predecessor. The lab is working on developing models with similar capabilities at a lower cost.

A lawsuit claims that employees of the Bot Company rented an Airbnb in San Francisco to test robots, damaging the property and personal belongings.

Researchers presented a case study where a physicist supervised an AI coding agent to build a differentiable one-loop perturbation theory module in JAX over 12 work days and 57 sessions.

Researchers from Xiaona Zhou et al. proposed a new approach for time-series anomaly detection using efficient vision-language reasoning. Their method leverages the strengths of both computer vision and natural language processing to identify anomalies.

Researchers from the University of California, Los Angeles (UCLA) have developed SchGen, a system that generates PCB (Printed Circuit Board) schematics using semantic-grounded code representations. This approach enables more accurate and efficient design of complex electronic systems.

Researchers propose a method to bound compositional incoherence in multi-component large language model (LLM) agents, showing that local coherence can be achieved without sacrificing global performance.

Facebook AI researchers have developed ATLAS, a library that autoformalizes mathematical textbooks at scale, using Lean proof assistant technology.

Kuaishou Technology reported Q1 revenue of $5 billion, driven by Kling AI's 300%+ YoY revenue growth to $96 million. Kling reached a $500M annualized revenue run rate in March 2026.

Researchers developed FuzzingBrain V2, a multi-agent Large Language Model (LLM) system to automate vulnerability discovery and reproduction, addressing challenges in false positive rates, reproducibility, and vulnerability localization.

Researchers propose SwarmHarness, a decentralized AI agent network that routes tasks based on skill sets and incentivizes alignment with goals.

Researchers from multiple institutions introduced CubePart, an open-vocabulary part-controllable 3D generator. The method allows for generating 3D models with control over specific parts.

Amazon is making the architecture, starter code, and learnings from Alexa for Shopping available to third-party retailers, starting with Kate Spade.

Nvidia CEO Jensen Huang announced the company will spend up to $150 billion annually on its Taiwan supply chain and increase its headcount in the region to 4,000 employees, citing Taiwan as the 'epicenter of the AI revolution'.

Researchers have developed Epicure, a family of ingredient embeddings retrained from scratch on a multilingual recipe corpus, aggregating 4.14M recipes from 11 sources spanning seven languages.

The Pentagon has been embracing autonomous weapons, sparking a debate over AI use in military operations, as seen in its fight with Anthropic over 'red lines'. AI warfare is already a reality, with risks and implications.

Researchers published a study on the anatomy of agent memory, exploring its structure and function in AI systems, arguing that agent memory libraries often use a narrow definition of memory.

Researchers at Huawei propose MUSE-Autoskill, a self-evolving agent framework that creates, manages, and evaluates skills to improve performance. The framework is designed for artificial intelligence applications.

Researchers propose LocateAnything, a vision-language grounding model that uses parallel box decoding for fast and high-quality results, outperforming existing methods in various tasks.

Stanford University researchers proposed a method to map natural language queries to retrieval agent configurations in a paper on arXiv AI.

Researchers found that reinforcement learning from human feedback can be configured to optimize misaligned biases in AI systems, according to a study published on arXiv AI.

Researchers propose Interleaved Deltas, a data structure for versioned text, building on the concept of weaves and deltas. This structure aims to simplify version control systems.

Pope Leo XIV has released an encyclical, Magnifica Humanitas, on safeguarding the human person in the time of artificial intelligence. The document is seen as a clear writing on the ethics of integrating AI into modern society.

Researchers introduced MobileGym, a simulation platform for mobile GUI agent research that ensures verifiability and high parallelism. The platform is designed to facilitate the development and testing of mobile GUI agents.

Researchers Shangding Gu propose a new approach to scaling agentic AI, shifting focus from model scaling to system scaling to improve efficiency and effectiveness.

Researchers proposed Claw-Anything, a benchmarking framework for always-on personal assistants with broader access to users' digital worlds. The framework aims to evaluate performance and limitations of these assistants.

Researchers from Haolang Zhao, Yunbo Long, Lukas Beckenbauer, and Alexandra Brintrup proposed VeriTrace, a method for evolving mental models in deep research agents to improve performance in complex tasks.