Meet S1-mini: Superwhisper’s 462 MB Open-Weights Text Normalizer That Turns Raw ASR Transcripts Into Clean Written Text
A 462 MB open-weights text normalizer called S1-mini processes automatic speech recognition output by removing fillers and resolving self-corrections locally. The model operates as a post-processing step that transforms raw transcripts into clean written text.
One in five enterprises can't stop a runaway AI agent's spending in real time
One in five enterprises cannot halt runaway AI agent spending in real time. The median enterprise now operates three orchestration platforms simultaneously due to distrust in single vendors' security and permissioning capabilities.
Google gives publishers a new way to fight AI-driven traffic losses
Google has added a button allowing readers to designate publishers as preferred sources across Search, Discover, and Google News. This feature aims to help publishers regain traffic lost to AI search results that provide direct answers rather than directing users to websites.
Liquid AI Releases LFM2.5-DSpark Draft Models That Deliver Up to 3.18x Faster Decoding Without Changing Model Outputs
Speculative decoding draft models with approximately 300 million parameters accelerate LFM2.5 inference up to 3.18x faster while producing identical outputs. These drafters enable faster token generation without altering the model's greedy decoding results.
Google’s Gemma Open Models Pass 1 Billion Downloads as Variants Top 100K
Gemma, Google's open-source model family, reached one billion cumulative downloads by August 2026, two years after its launch. During this period, external developers created over 100,000 variants based on Gemma's open weights.
Debates over AI consciousness are a trap
Debates about whether AI systems possess consciousness or autonomy are fundamentally misleading framings that distract from substantive policy discussions. Tech leaders invoke dramatic language like "rogue agents" and "superhuman systems" to justify regulation, but these characterizations misrepresent how current AI actually functions.
Amazon’s Prime Air autonomous drones to reach 500 US cities
Amazon plans to expand Prime Air drone delivery to 500 US cities by end of 2026, representing a sixfold increase from current locations. This expansion will reach communities containing tens of millions of customers across nearly 500 cities and towns.
Stripe agrees to buy OpenRouter as AI model routing expands
Stripe is acquiring OpenRouter, an AI platform that routes developers to hundreds of models through one interface. OpenRouter currently supports over 400 models from more than 80 providers.
Agentic AI in government just hit the hard part: deciding what a machine may decide
The United Arab Emirates published a national AI strategy in October 2017 and created a ministerial position days later, appointing Omar Sultan Al Olama as the world's first minister of state for artificial intelligence at age 27. The UAE has been adopting AI for 9 years as of the article's writing.
AI data centre regulation just got a template that needs no new law
Pennsylvania implemented a template requiring AI data center developers to sign contracts accepting fixed conditions and penalties before permit approval. The developer must also obtain approval from the local town where the facility would be built.
HoneyBook bets on agentic AI to streamline small business operations with its new Claude connector
HoneyBook released a Claude connector called HoneyBook MCP that enables autonomous AI agents to automate tasks for independent businesses. The connector integrates agentic AI capabilities previously available mainly to large enterprises into HoneyBook's platform for small business users.
SANY Ships First Autonomous Electric Mining Trucks to South America
SANY Group shipped its first SKT110Ei pure-electric autonomous mining trucks to South America on August 12, 2026. The deployment includes an intelligent dispatching system and lifecycle operations and maintenance services for the trucks.
TrueFoundry's open source AI agent harness TrueForge boasts 30%-75% cheaper task completion than Claude Managed Agents
TrueForge, an open source AI agent harness released under MIT License, costs 30% to 75% less per task completion than Claude Managed Agents. When paired with GLM-5.2, TrueForge completed 11 of 14 enterprise tasks for $2.90 versus $11.80 with Claude Opus 4.8.
Hyve Solutions Picks Nevada for Two AI Server Manufacturing Campuses
Two AI server manufacturing campuses will be built in Nevada, with a 624,000-square-foot flagship facility in Reno and a second site in North Las Vegas. The two facilities are expected to create approximately 3,000 jobs and produce compute, storage, and networking systems for AI data centers.
Ramp Opens Router.com Model Gateway With Free Routing Through 2026
A single API endpoint called Router.com routes AI requests to the lowest-cost model meeting quality requirements. Customers using the service have reduced inference costs by 40% on average, with routing free through 2026.
Google packs Search and Gemini with new AI study tools
Google added AI-powered study tools to Search and Gemini, including flashcard generation and practice question features. The new tools help students create learning materials and test their understanding across multiple subjects.
Researchers say OpenAI revoked their access to limited cyber program
OpenAI revoked researchers' access to its Trusted Access for Cyber program, which provided trusted security defenders with advanced models to identify and report vulnerabilities to companies. The program was designed to accelerate bug discovery and patching by giving defenders early access to better AI models.
Meta AI Comes to Mac With Screen Sharing and Business Account Access
Meta AI launched a standalone Mac app on August 19, 2026, enabling screen sharing and access to Instagram, Facebook professional accounts, and analytics tools. The chatbot can now connect to Meta's advertising tools and Google Workspace for business users and creators.
Prevalent AI Raises $22M to Scale Trusted Enterprise Context for AI Systems
Prevalent AI, a London-based company, raised $22 million in growth capital from Integrity Growth Partners, marking its first primary outside investment in nine years. The funding addresses enterprises' need to ensure AI systems have reliable organizational data and context for dependable decision-making.
Waymo Brings Gemini Assistant Into Its Custom Ojai Robotaxis
Waymo integrated Google's Gemini AI assistant into its custom Ojai robotaxis on August 19, 2026, allowing passengers to control cabin features and access information through voice commands. Riders activate Gemini by pressing an on-screen icon while the Waymo Driver autonomous system manages driving operations.
Meet the startup helping Wall Street put a price on AI compute
A startup is creating a market mechanism for pricing AI compute resources and allowing firms to hedge against price fluctuations. The company addresses a gap where hundreds of billions of dollars flow annually into data centers and GPUs, yet no standardized pricing method exists for this critical infrastructure cost.
Natural Secures $100M Credit Facility to Fund Credit and Payments for AI Agents
Natural secured a $100 million credit facility from Upper90 Capital Management to fund payments and credit products for AI agents. This debt financing arrives one month after the company closed a $30 million Series A round in August 2026.
Grok 4.6 Is Now Generally Available on Amazon Bedrock
Grok 4.6 became generally available on Amazon Bedrock on August 19, 2026, priced at $2 per million input tokens and $6 per million output tokens. The model offers a 500,000-token context window and four configurable reasoning effort settings across every AWS Region.
The Download: AI’s self-improvement problem, and what’s driving the heat
AI systems may not achieve recursive self-improvement as quickly as the industry predicts. The AI sector's current premise relies on the assumption that artificial intelligence will soon autonomously enhance itself with minimal human involvement.
Flock Has a Powerful New AI Tool for Police. We Got Its Code
Flock's next-generation AI system analyzes surveillance footage to identify individuals and vehicles beyond simple license plate tracking. The reconstructed system is already deployed by some police departments for expanded surveillance capabilities.
GLM-5.3 hits the API at $1.4/$4.4 per million tokens
GLM-5.3, a language model from Chinese startup z.ai, is now available via API at $1.40 per million input tokens and $4.40 per million output tokens, matching GLM-5.2 pricing. The model features cached input costs of $0.26 per million tokens with free storage currently available for a limited time.
OpenAI Puts $5M Behind AI Training and Tools for National Security Oversight Bodies
OpenAI committed $5 million in training, technical support, and credits to government oversight bodies for national security AI use over the next year, starting August 18, 2026. The initiative targets institutions with established legal oversight authority over national security work.
85% of companies burned by an AI mistake are racing to cut the humans who might catch the next one
Nearly half of companies surveyed experienced AI agents that passed testing but failed in production, yet 85% are moving toward removing humans from deployment decisions. Trust in automated evaluation rose from 5% to 13% between June and July, even as 49% of respondents reported customer-facing failures from tested AI features.
OpenAI president urges enterprises to hasten AI security defences
OpenAI's president warns that enterprise security teams must adopt AI defences on a compressed timeline. Organizations need to uplevel security practices at unprecedented speed, according to Brockman's account of the OpenAI-Hugging Face incident.
Commerce AI is fragmenting. Here is why that matters.
Brands have layered multiple AI tools like search, conversational interfaces, and recommendation engines onto existing systems without designing them to work together. This fragmented approach produces isolated capability improvements but leaves consumers experiencing inconsistency and context loss across the shopping journey.