AI News — Saturday, June 13, 2026

9
EvoArena: Tracking Memory Evolution for Robust LLM Agents in Dynamic Environments

Researchers introduce EvoArena, a new benchmark and framework for evaluating how LLM agents evolve and manage memory to maintain robustness in constantly changing environments.

Hugging Faceresearch
9
Meta's Months-Old AI Unit Described as a 'Soul-Crushing Gulag' by Engineers

Engineers within Meta's recently formed AI unit reportedly describe their working conditions as extremely challenging and demoralizing, raising concerns about internal culture and project pressures.

TechCrunchindustry
9
Google Sues Chinese Cybercrime Operation Using AI to Scam Hundreds of Thousands

Google has filed a lawsuit against a Chinese cybercrime group accused of leveraging AI to defraud hundreds of thousands of victims, highlighting the growing threat of AI-powered scams.

TechCrunchindustry
8
MiniMax Sparse Attention: Efficient Attention for Large Language Models

A new sparse attention mechanism, MiniMax Sparse Attention, is proposed to improve the efficiency of large language models by focusing on the most critical tokens.

Hugging Faceresearch
8
SpatialClaw: Rethinking Action Interface for Agentic Spatial Reasoning

SpatialClaw presents a novel action interface designed to enhance the spatial reasoning capabilities of AI agents, allowing for more intuitive and effective interaction with 3D environments.

Hugging Faceresearch
8
BBVA Integrates OpenAI to Place AI at the Core of Banking Operations

BBVA is partnering with OpenAI to embed AI deeply into its banking services, aiming to revolutionize financial operations and customer experiences.

OpenAI Blogindustry
7
InterleaveThinker: Reinforcing Agentic Interleaved Generation

InterleaveThinker introduces a method for reinforcing interleaved generation in AI agents, enabling them to better combine planning and execution for complex tasks.

Hugging Faceresearch
7
Robust-U1: Can MLLMs Self-Recover Corrupted Visual Content for Robust Understanding?

This paper investigates the ability of Multimodal Large Language Models (MLLMs) to self-recover and robustly understand visual content even when it is corrupted.

Hugging Faceresearch
7
FORT-Searcher: Synthesizing Shortcut-Resistant Search Tasks for Training Deep Search Agents

FORT-Searcher proposes a method to create challenging, shortcut-resistant search tasks, crucial for training more robust and intelligent deep search agents.

Hugging Faceresearch
7
MaxProof: Scaling Mathematical Proof with Generative-Verifier RL and Population-Level Test-Time Scaling

MaxProof introduces a new approach to scale mathematical proof generation using a combination of generative-verifier reinforcement learning and population-level test-time scaling.

Hugging Faceresearch
7
OpenAI Launches New Academy Courses for Applying AI in the Workplace

OpenAI has introduced new Academy courses designed to equip professionals with the skills needed to effectively apply AI technologies in various work environments.

OpenAI Blogproduct
6
LabVLA: Grounding Vision-Language-Action Models in Scientific Laboratories

LabVLA introduces a new framework for training vision-language-action models specifically designed to operate and assist within complex scientific laboratory environments.

Hugging Faceresearch
6
I Switched to the Agent Toolkit for AWS: Here's Why

An AWS developer shares their positive experience and reasons for adopting the Agent Toolkit for AWS, highlighting its benefits for building AI agents.

Dev.toproduct
6
I Lead AI Agents Every Day - Here Are 5 Shifts No Standard Tells You How to Make

An experienced AI agent leader shares five crucial, non-standard shifts in approach necessary for effectively managing and optimizing AI agents in daily operations.

Dev.toindustry
6
EurekAgent: Agent Environment Engineering is All You Need For Autonomous Scientific Discovery

EurekAgent proposes that effective environment engineering for AI agents is the key to achieving fully autonomous scientific discovery.

Hugging Faceresearch