All Updates
Page 48 of 1371
June 18, 2026
Circles Spy Tools Sold to Repressive Regimes
A Bulgaria-based company, Circles, is under scrutiny for selling surveillance technology to governments with records of repression. The tools reportedly enable mobile phone tracking and eavesdropping on private communications.
Xcientist: A Research Harness for Accountable AI Science
Xcientist is a new research harness designed to externalize AI scientific workflows into inspectable, contract-governed processes. It tracks the entire research lifecycle, from literature synthesis to experimental validation, to prevent claim drift and ensure scientific accountability.
WorldLines: Benchmarking Long-Horizon Stateful Embodied Agents
WorldLines is a new benchmark designed to evaluate embodied agents on long-term memory and household task planning. It introduces the ObsMem framework to address challenges in partial observability and state tracking for complex, extended interactions.
The Rise of Predatory OnlyFans Management Agencies
An investigative report into the emergence of 'OnlyFans managers' who exploit creators through aggressive scaling tactics and predatory revenue-sharing models. The article highlights how these middlemen use lifestyle-based marketing to recruit and manipulate young creators.
R2D-RL: Bridging RoboCup Soccer and Modern Python MARL
R2D-RL is a new environment that integrates the RoboCup 2D Soccer Simulation (RCSS2D) with modern Python-based multi-agent reinforcement learning (MARL) workflows. It provides a synchronized, high-performance interface for training agents in complex, adversarial soccer scenarios.
Optimizing Lithium Production via POMDP Decision Framework
This research introduces a Partially Observable Markov Decision Process (POMDP) framework to optimize lithium mining decisions under geological, demand, and pricing uncertainties. The model outperforms human heuristics by dynamically adapting to shifting market conditions and technology choices.
Optimizing Human-AI Team Coordination for Better Performance
This research explores how shared-workspace human-AI teams coordinate tasks, finding that performance often suffers without proper structural scaffolding. By implementing group memory and human-in-the-loop gates, teams can significantly improve their collaborative efficiency.
ForecastBench-Sim: A Simulated-World Forecasting Benchmark
ForecastBench-Sim is a new benchmark for AI forecasting that uses Freeciv game simulations to overcome real-world data limitations. It enables researchers to test probabilistic reasoning on dynamic, immediately resolvable, and counterfactual scenarios.
First In-Orbit Zero-Shot Vision-Language Model Demonstration
NAVI-Orbital successfully demonstrated the first in-orbit autonomous multi-modal inference using a vision-language model on a LEO satellite. The system utilizes Gemma 3 to classify imagery and respond to natural-language queries, enabling semantic compression of Earth observation data.
DeFAb: A Verifiable Benchmark for Defeasible Abduction in AI
DeFAb is a new benchmark designed to test foundation models on defeasible abduction, using formal logic to evaluate creativity and theoretical reasoning. It reveals that frontier models struggle with logical rigor, often failing to internalize defeasible reasoning compared to symbolic solvers.
CEO-Bench: Can AI Agents Play the Long Game?
CEO-Bench is a new benchmark that evaluates AI agents on their ability to manage a startup over a 500-day simulated period. It tests long-horizon planning, noisy data analysis, and adaptive decision-making in complex business environments.
CaVe-VLM-CoT: An Interpretable Vision-Language Model Framework
CaVe-VLM-CoT is a new agentic-RAG framework designed to reduce hallucinations in Vision-Language Models through a five-stage closed-loop verification pipeline. It introduces CaVeScore, a comprehensive metric for evaluating retrieval quality, citation faithfulness, and cross-modal grounding.
Shanghai clarifies IPO path for AI model developers
The Shanghai Stock Exchange has provided clear guidelines for unprofitable AI model developers to list on the Star Market. This move aims to help Chinese LLM firms secure necessary capital to compete with US-based AI labs.
visionOS 27 Developer Beta: New Environments and UI Features
The first developer beta of visionOS 27 introduces new immersive environments and curved window designs. These updates aim to enhance spatial computing interactions for developers and users.
Korea Rejects Antitrust Settlements for Delivery Apps
South Korea's antitrust regulator has rejected settlement bids from Baedal Minjok and Coupang Eats. Both companies face potential fines for alleged unfair business practices.
Midjourney expands into hardware with full-body ultrasound scanner
Midjourney CEO David Holz has unveiled 'The Midjourney Scanner,' a new hardware device that uses ultrasound sensors to create detailed internal body images. The company aims to provide image quality comparable to MRI scans for frequent health monitoring.
How to select the optimal AI model for business
Nomura Research Institute (NRI) expert Yuki Kitamura argues that business AI selection should not rely solely on benchmarks. A holistic approach considering specific business requirements is essential for success.
Hutong Research on PBOC Policy and Yuan Outlook
Guo Shan from Hutong Research analyzes potential shifts in China's central bank interest rate framework. The discussion covers the broader implications for the yuan's valuation in the current economic climate.
Fed Expected to Keep Interest Rates Steady
Betsey Stevenson discusses the outlook for the US economy, suggesting the Federal Reserve will maintain current interest rates. The analysis focuses on economic stability and future policy direction.
12306 Challenges OTA Platforms in Travel Market
Chinese regulators are cracking down on OTA platforms for misleading ticket sales, while the official railway platform 12306 is expanding into hotel and travel services to capture direct traffic.