AI News — Tuesday, August 18, 2026
Anthropic, a leading AI company, has reportedly achieved an astounding $65 billion in annualized revenue, signaling massive growth and market demand for its AI models.
Reports indicate that Amazon, which began as an online bookseller, is now allegedly destroying shipments of rare texts to use their content for training its artificial intelligence models.
A systematic evaluation explores the effectiveness of current detectors and generators against AI-generated video attacks targeting real-world crisis events, highlighting the challenges of social dissemination.
Nous Research has launched NousCoder-14B, a new open-source coding model positioned to compete with proprietary solutions like Claude Code, marking a significant development in open-source AI for developers.
OpenAI has published a new set of policy ideas aimed at guiding the development and deployment of AI in the evolving 'Intelligence Age,' addressing societal and governance challenges.
AI automation startup Relay has ceased operations, with its team being acquired by Google to join the Chrome division, indicating a trend of talent consolidation in the tech giant.
A new research paper introduces Self-Supervised Visual On-Policy Distillation, a method designed to enhance the learning efficiency and performance of AI agents through visual self-supervision.
OpenAI details 'The Defender’s Window,' outlining its strategies and initiatives for ensuring the safety and security of its AI systems against potential misuse and vulnerabilities.
Researchers introduce Intern-S2-Mobius, a novel foundation model architecture that achieves advanced performance by decoupling knowledge acquisition from reasoning capabilities.
This article argues that the primary risk in using AI for coding lies not in the AI's ability to generate code, but in developers failing to thoroughly understand and verify the AI-produced output.
New research suggests that large language models are spontaneously developing modular cognitive architectures, indicating a more structured and potentially interpretable internal organization.
DFM Mimir v1 is presented as an open-source Human-Robot Interaction (HRM) model that achieves frontier performance with only 1 billion parameters, utilizing exclusively permissible post-training data.
MobileMem explores the potential of training AI models by leveraging a full year's worth of diverse mobile user experiences, aiming to create more context-aware and helpful mobile AI.
Marionette introduces a new generative AI model capable of predicting complex world states, rendering detailed 3D geometry, and painting realistic appearances, pushing the boundaries of virtual world creation.
SimpleOPD proposes a novel tokenizer-agnostic on-policy distillation method designed to improve long-context reasoning capabilities in language models, simplifying the training process.