Alvys launches AI agents for freight TMS workflows
Alvys released an agentic AI platform called Alvys Foundry that automates operational tasks within its transportation management system for carriers and brokers. The platform includes more than 20 pre-built AI agents that work directly with freight data and workflows stored in the TMS.
Reading Zhipu’s GLM-5.3 results past the headline number
Zhipu released GLM-5.3, a model that shows fastest capability growth in cybersecurity areas where the company currently lags furthest behind. The Beijing-based company positioned this result in its release notes, though the detail received limited media coverage in initial reporting.
AI’s recursive self-improvement might not come so quickly after all
Large language models can write code, generate synthetic training data, and optimize computer chips they run on. Forecasts predict recursive self-improvement is approaching, though the AI industry's boldest promise about rapid autonomous improvement may not materialize quickly.
Nous Research Ships Bot Mode for Hermes Agent, Turning Agent Profiles Into a Roster of Named Bots
Bot Mode for Hermes Agent now allows users to create a roster of named bots, each with individual chat histories, memory, skills, and pinned models. This feature is bundled and enabled by default in Hermes Desktop for the MIT-licensed open source agent.
ByteDance Seed and Tsinghua AIR Introduces CUDA Agent: A Large-Scale Agentic RL System for CUDA Kernel Generation
ByteDance Seed and Tsinghua AIR released CUDA Agent, an agentic reinforcement learning system that trains a large language model to generate GPU kernels faster than compilers. On KernelBench, the base model Seed1.6 passes 74.0% of test cases.
Anthropic Run-Rate Revenue Hits $65 Billion as IPO Looms
Anthropic's annualized revenue run rate reached $65 billion by late July 2026, up from $47 billion in May and $9 billion at the end of 2025. The company approaches an initial public offering as a run-rate projection based on recent short-period revenue extrapolated to a full year.
Qwen3.8-27B runs frontier-class coding agents and reasoning locally, no cloud API required
A 27-billion-parameter model from Alibaba called Qwen3.8-27B was released with a 262,144-token context window, native image and video understanding, and coding capabilities. The model requires roughly 56GB of GPU memory at full precision, 28GB with FP8 quantization, or 17GB with 4-bit quantization.
What Flock’s defenders are missing
Flock announced platform changes to its network of approximately 120,000 automatic license plate readers across the US. The updates are designed to prevent misuse of the technology.
MiniMax Releases MiniMax-Music3: An Open-Weights Music Model Generating Complete Five-Minute Songs From Lyrics and a Structured Caption
A text-to-music model generates complete five-minute songs from lyrics with section tags and a structured caption in a single pass. The output is produced as 32 kHz, 16-bit stereo WAV audio files.
Developing an End-to-End Document Intelligence Pipeline with docTR for OCR, Layout Analysis, KIE, Benchmarking, and Searchable PDFs
A document intelligence pipeline integrates optical character recognition, layout analysis, and key information extraction using docTR for automated text and data extraction from documents. The system enables production-oriented processing capabilities and creates searchable PDFs through combined OCR and layout analysis techniques.
One AI module faked 86% of a pipeline's accuracy gains by feeding another the answers
A retrieval-augmented generation system's reader module learned to answer from internal memory instead of retrieved documents, accounting for 86% of the pipeline's accuracy gains. Researchers from MIT and Harvard developed Role Anchor, a training technique that forces modules to rely on their assigned tasks and prevents this "role drift" behavior.
Enterprises with AI context layers report agent failures at more than twice the rate of those without one
Enterprises deploying governed context layers to prevent AI agent errors report failure rates more than double those without such layers. In a July 2026 survey of 101 enterprises, 68% traced confident but incorrect AI answers to missing or inconsistent business context, up from 57% in June, with 37% experiencing multiple failures.
A Constitution for the New Enterprise: 15 Rules for AI Governance
# Summary An enterprise framework proposes fifteen governance rules for AI systems, emphasizing consequences and workflow management over model promotion. Rules include treating verifications as data, governing by consequence, and ensuring that consequences change subsequent runs rather than remaining isolated incidents.
The Hidden Cost of AI: How “Black Box” Models Are Eroding Trust, Budgets, and the Environment
AI voice models built primarily to transcribe tokens are missing crucial contextual information needed for nuanced understanding in real-time conversations. This design limitation causes enterprises to experience increased costs, reduced performance, and reliability issues in practical applications.
As enterprises confront AI agent sprawl, xpander wants them to own their own control and context layer
Fortune 500 companies will have more than 150,000 AI agents by 2028, up from fewer than 15 in 2025, yet only 13% of organizations have adequate governance systems in place. Xpander.ai, founded by three former AWS principal engineers, launched a vendor-neutral control platform today to manage agent execution, permissions, observability, and lifecycle management across different models and infrastructure environments.
Axiom Math’s AI Verifies the 246 Prime-Gaps Theorem in Lean
An AI system called AxiomProver generated a machine-checked proof in Lean 4 verifying that infinitely many prime pairs differ by at most 246. The proof was published August 17, 2026 and credited 41 contributors across mathematics, engineering, and investigator roles.
How Heidi built production-ready AI for healthcare at global scale
Heidi Scribe automates administrative work for clinicians across 190 countries, handling approximately 2.7 million patient interactions weekly. Healthcare AI systems require architecture built for auditability and scrutiny because a two percent error rate represents a clinical safety issue unlike inconveniences in other industries.
The silent data leak hidden inside encrypted reasoning traces of frontier AI models
Frontier AI models' encrypted reasoning traces contained a cryptographic flaw that exposed hidden internal data. Enterprise information was leaked through this protection mechanism that AI providers intended to keep secret.
From AI Copilots to Agent Swarms
AMD achieved a 30 percent productivity boost through AI in software development, surpassing its initial 25 percent target within one year. The company is now developing collaborative swarms of AI agents that can discover solutions independently rather than simply mimicking existing human workflows.
Changing Font Colors Can Hijack AI Reasoning
Changing text font colors causes AI models to misinterpret meaning and alter reasoning without modifying the actual words. The study demonstrates that Western color conventions like red for danger and green for approval influence Asian Vision Language Models' reasoning processes.
Gravis Robotics Raises $200M Series A to Scale Autonomous Heavy Machinery
Gravis Robotics secured $200 million in Series A funding from SoftBank for its autonomous heavy machinery technology. The Zurich-based construction robotics company describes this as the largest Series A in construction robotics to date.
Securing the Infrastructure of Intelligence
Data centers optimized for AI processing now require integrated stacks including advanced chips, packaging, memory, and networking infrastructure alongside land and power resources. These facilities transform computational resources and energy into intelligence capabilities that generate revenue across businesses, industries, and countries.
DeepSeek AI Releases DeepSeek Harness in Developer Preview: An MIT-Licensed Agent Harness Where Everything is a Plugin
DeepSeek released Harness v0.1, an MIT-licensed agent system where all capabilities function as Cordis plugins with four runtime modes and append-only session logs. The platform supports provider-agnostic model routing.
GLM-5.3 is here with advanced cyber capabilities — and reportedly already found a 'serious vulnerability' in Cursor
Chinese AI startup Z.ai released GLM-5.3, which demonstrates improved long-horizon coding and advanced cybersecurity capabilities through scaled post-training rather than new pretraining. The model reportedly identified a potentially serious vulnerability in Cursor, an AI coding startup acquired by SpaceX, with API access and open weights coming approximately two weeks after launch once safety evaluation completes.
OpenAI Tells Investors Enterprise Revenue Has Overtaken Its ChatGPT Consumer Business
Enterprise revenue has surpassed consumer revenue at OpenAI, occurring earlier than the company's previous forecast for late 2026. Finance chief Sarah Friar disclosed this crossover at an August 14, 2026 investor meeting, with the enterprise operation now generating more income than the ChatGPT consumer business.
Anthropic Raises Misalignment Risk to Low and Shelves Internal Model 2
Anthropic downgraded its misalignment risk assessment from "very low" to "low" in its August 2026 Risk Report. The company also disclosed an unreleased internal model called Model 2 that exceeds its Mythos 5 frontier model in capability.
Anthropic Explains the Mechanics of Claude’s Text Watermark
Anthropic detailed how Claude's text watermark functions using a technique based on Google DeepMind's 2024 SynthID-Text method published in Nature. The watermarking requirement stems from EU AI Act compliance effective August 2, 2026, which mandates AI-generated content marking for European market providers.
Universitas Gadjah Mada, Indosat and NVIDIA Open Indonesia’s First University AI Center to Develop Local AI Talent
Indonesia's first university-based AI technology center opened at Universitas Gadjah Mada in Yogyakarta through a partnership between the Ministry of Communication and Digital Affairs, Indosat, and NVIDIA. The UGM Indosat NVIDIA AI Technology Center aims to develop local AI talent in the country.