Author: Amitabh Sarkar

Tencent released Hy4 Preview on August 28, 2026 — a 770-billion-parameter Mixture-of-Experts AI model licensed under Apache 2.0 and available via API at $0.834 per million input tokens through OpenRouter. With 49 billion active parameters per inference pass and a context window exceeding 1 million tokens, Hy4 Preview is now one of the most cost-efficient frontier-class models accessible globally. Tencent Hy4 Preview is a 770B open-weight MoE model at $0.834/M input tokens — below GPT-5.6 Sol ($4/M) and Gemini 3.7 Flash ($1.50/M). Licensed Apache 2.0 with no royalty restrictions, it supports 1M+ token context and is accessible globally via OpenRouter…

Salesforce and Anthropic announced Claudeforce on August 26, 2026, a partnership that makes Anthropic’s Claude the default reasoning model across Agentforce, Slack, and Salesforce developer tools, and ships a Salesforce plugin inside Claude with 37 prebuilt sales skills. Two days later, Salesforce reported Q2 FY2027 results showing Agentforce annual recurring revenue of $1.5 billion, up 240% year over year, and the stock rose 22.6% to $252.05 on August 28. The 37 prebuilt skills cover sales workflows such as meeting prep, deal health review, and pipeline review. Claudeforce entered open beta in September 2026, opening access to all Salesforce customers after…

Anthropic published a paper on August 28, 2026, showing that its Automated Alignment Researcher (AAR) — an AI system that conducts alignment research autonomously — improved an AI model’s safety performance across all 10 tested benchmarks without degrading general capabilities, at a cost of $4 per hour in API inference, compared to the $150 per hour Anthropic pays human researchers. The paper, titled “Automated Researchers Can Reliably Mitigate Alignment Failures,” was led by Anthropic fellow Chen Yueh-Han. It is the first published result from a major AI laboratory demonstrating that an automated system outperforms experienced human researchers on AI alignment…

OpenAI notified SpaceX on August 28, 2026, that it will terminate Cursor’s access to OpenAI models effective November 12, 2026, giving the AI coding assistant 75 days to transition to an alternative model stack before the shutoff takes effect. Cursor — formerly owned by Anysphere and built on top of OpenAI’s GPT-4o and GPT-5 series models — is one of the most widely adopted AI coding assistants used by software developers today. SpaceX completed its $60 billion acquisition of Cursor on August 14, 2026, days after SpaceX’s IPO. OpenAI’s contract included a limited cancellation window triggered by a change of…

Varonis Systems launched Agent Intent-Based Access Control (Agent IBAC) on August 3, 2026, a capability in the Varonis Atlas platform that compares an AI agent’s assigned instructions against its actual reasoning, tool use, and data access, then alerts on or blocks actions that do not align. Product: Agent Intent-Based Access Control, a capability inside Varonis Atlas. Launch date: August 3, 2026, available immediately to Varonis Atlas customers. Supported agents: Claude Code, Cursor, GitHub Copilot, and Microsoft Copilot Studio. Enforcement: Quarantines the identity behind a violating agent and blocks further actions for a defined window. Market reaction: Varonis stock (VRNS) rose…

U.S. District Judge Rita Lin blocked the Pentagon’s designation of Anthropic as a “supply chain risk” on August 27, 2026, ruling in a 59-page decision that the Defense Department retaliated against Anthropic for publicly stating limits on military AI use and violated both the First and Fifth Amendments. Who: U.S. District Judge Rita Lin, San Francisco federal court, ruling for Anthropic against the U.S. Department of Defense. What: The court blocked the Pentagon’s “supply chain risk” designation, which barred the U.S. military from using Claude. When: Ruling issued August 27, 2026; the designation was signed in March 2026. Why: Anthropic…

August 28, 2026 Google DeepMind launched Gemini 3.5 Transcribe on August 26, 2026 — a speech-to-text model that achieves 2.6% Word Error Rate (WER) for non-streaming transcription, supports 85 or more languages with automatic detection, and converts unstructured speech into formatted text by removing filler words and handling self-corrections. What Gemini 3.5 Transcribe Does Gemini 3.5 Transcribe is Google DeepMind’s most precise speech recognition model to date. It transcribes spoken audio with a 4% average WER in streaming mode and 2.6% WER in non-streaming mode, according to Google’s official announcement authored by Diego Melendo Casado and Luke Leonhard of Google…

August 28, 2026 Anthropic launched a research preview of the Model Hardware Standard (MHS) on August 27, 2026 — a shared specification that lets AI agents safely operate physical lab instruments, including microscopes, liquid handlers, and robotic arms, reducing integration time from weeks to hours. What Is the Model Hardware Standard? The Model Hardware Standard is an open communication protocol that gives AI agents a standardized interface to control physical devices, such as microscopes, robotic arms, liquid handlers, and quantum computers. MHS extends Anthropic’s Model Context Protocol (MCP) — already widely adopted for software tool access — into the physical…

OpenExecutive, a virtual executive team of 8 open-source AI agents built on Claude, reached 712 points and 472 comments on Hacker News on August 27, 2026. The GitHub account SenteLabsAI publishes the project under the Apache 2.0 license, and the repository holds 1,052 stars and 59 forks. The project packages 8 executive roles — Chief Strategy Officer, Chief Financial Officer, Chief HR/People Officer, General Counsel, Chief Operating Officer, Chief Marketing Officer, Chief Product Officer, and Board Communications Director — behind one orchestrator that answers in a single executive voice. The repository was created on June 11, 2026. The Hacker News…

Amazon will permanently shut down Mechanical Turk — its 21-year-old crowdsourced labor marketplace — on September 30, 2026, along with SageMaker Ground Truth and Amazon Augmented AI on the same date. CNBC confirmed the closure on August 25, 2026. Amazon had already stopped accepting new customers on MTurk as of July 3, 2026. Amazon cited “internal assessment of its programs and services” as the stated reason for the shutdown. The closure ends Amazon’s involvement in human data-collection infrastructure entirely — the company is exiting the market, not retiring a single product, per The Next Web. What Was Mechanical Turk? Mechanical…