<?xml version="1.0" encoding="UTF-8"?>
<feed xmlns="http://www.w3.org/2005/Atom" xml:lang="en">
  <id>https://theaisift.com/</id>
  <title>The AI Sift</title>
  <subtitle>The curated AI read for founders who can't fall behind — every story ends with the decision it puts on your desk.</subtitle>
  <link href="https://theaisift.com/" rel="alternate" type="text/html"/>
  <link href="https://theaisift.com/feed.xml" rel="self" type="application/atom+xml"/>
  <link href="https://theaisift.com/feed.xml" rel="hub"/>
  <updated>2026-07-29T06:36:55.247Z</updated>
  <author>
    <name>Clements Emerson Dewanto</name>
    <email>hello@theaisift.com</email>
  </author>
  <icon>https://theaisift.com/favicon.svg</icon>
  <logo>https://theaisift.com/og.png</logo>
  <generator>The AI Sift</generator>
  <rights>© 2026 YOU-DO-NOTHING</rights>
  <entry>
    <id>https://theaisift.com/news/cyera-acquires-oasis-security-1b-2026</id>
    <title>Cyera Acquires Oasis Security for $1B as Non-Human Identity Becomes a Platform War</title>
    <link href="https://theaisift.com/news/cyera-acquires-oasis-security-1b-2026" rel="alternate" type="text/html"/>
    <published>2026-07-29T12:00:00.000Z</published>
    <updated>2026-07-29T12:00:00.000Z</updated>
    <summary type="html">Data security unicorn Cyera signed a letter of intent to acquire non-human identity startup Oasis Security for $1 billion, its third acquisition this year and the clearest signal yet that governing what AI agents can access is becoming a standalone buying category rather than a feature of existing identity platforms.</summary>
    <category term="Capital"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/anthropic-amodei-open-weights-position-2026</id>
    <title>Dario Amodei Breaks His Silence. Anthropic Has Never Called for an Open-Weight Ban.</title>
    <link href="https://theaisift.com/news/anthropic-amodei-open-weights-position-2026" rel="alternate" type="text/html"/>
    <published>2026-07-28T12:00:01.000Z</published>
    <updated>2026-07-28T12:00:01.000Z</updated>
    <summary type="html">Anthropic CEO Dario Amodei published a blog post Monday explicitly stating his company has &quot;never advocated for a ban on open-weights models,&quot; after four days of industry speculation that Anthropic was lobbying for exactly that. He agreed with much of the open-weight letter but disputed its safety claims, and laid out three policy measures he supports instead: keeping powerful chips out of China, cracking down on industrial-scale distillation, and mandatory safety testing for all sufficiently capable models.</summary>
    <category term="Policy"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/claude-shared-chats-google-indexed-2026</id>
    <title>Anthropic never shipped a noindex tag. Hundreds of Claude shared chats landed on Google.</title>
    <link href="https://theaisift.com/news/claude-shared-chats-google-indexed-2026" rel="alternate" type="text/html"/>
    <published>2026-07-28T12:00:00.000Z</published>
    <updated>2026-07-28T12:00:00.000Z</updated>
    <summary type="html">Anthropic&apos;s Claude share feature produced public URLs with no noindex meta tag, allowing Google and Bing to crawl and index hundreds of shared conversations and Artifacts containing API keys, resumes, legal notes, medical histories, and apparent SSNs. Anthropic patched robots.txt around July 27 but Bing and third-party scrape sites were still serving results.</summary>
    <category term="Agents &amp; Tools"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/open-weight-letter-bandwagon</id>
    <title>90 Companies Now. None of Them Had a Choice.</title>
    <link href="https://theaisift.com/news/open-weight-letter-bandwagon" rel="alternate" type="text/html"/>
    <published>2026-07-27T12:00:00.000Z</published>
    <updated>2026-07-27T12:00:00.000Z</updated>
    <summary type="html">The open-weight letter published July 24 added more than 65 signatories in 72 hours, growing from 25 companies to more than 90. SpaceX joined the list after Elon Musk initially voiced support without signing. Anthropic and Amazon, the two most visible holdouts, issued no public response. The story is no longer who signed. The story is that not signing became impossible.</summary>
    <category term="Policy"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/open-weight-letter-white-flag</id>
    <title>The Open-Weight Letter Is a White Flag Disguised as a Manifesto</title>
    <link href="https://theaisift.com/news/open-weight-letter-white-flag" rel="alternate" type="text/html"/>
    <published>2026-07-26T12:00:00.000Z</published>
    <updated>2026-07-26T12:00:00.000Z</updated>
    <summary type="html">On July 24, 2026, 25 companies signed an open letter to Washington defending open-weight AI. Jensen Huang created an X account just to post it. The most revealing story is who didn&apos;t sign: Anthropic, Amazon, and xAI. The deeper story is that the letter exists because the Trump administration is considering broad restrictions on Chinese open-weight models like Kimi K3, DeepSeek, and Qwen, and US industry is terrified of what a wide ban would do to their own businesses, not China&apos;s. The unstated argument: restricting open-weight to hurt China helps China win the global diffusion war.</summary>
    <category term="Policy"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/klaimee-5m-seed</id>
    <title>Klaimee Raises $5.5M Seed to Insure What Cyber and E&amp;O Policies Won&apos;t Touch: Autonomous AI Agents</title>
    <link href="https://theaisift.com/news/klaimee-5m-seed" rel="alternate" type="text/html"/>
    <published>2026-07-25T12:00:01.000Z</published>
    <updated>2026-07-25T12:00:01.000Z</updated>
    <summary type="html">Klaimee, a San Francisco InsurTech startup, raised $5.5 million in seed funding to provide insurance-backed performance warranties for autonomous AI agents. Founded by ex-SafetyWing operator Ines Boutemadja and engineer Julien Catonnet, the company fills a coverage gap that neither cyber insurance nor tech E&amp;O policies were designed to address.</summary>
    <category term="Capital"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/claude-opus-5</id>
    <title>Anthropic ships Claude Opus 5 at half Fable 5&apos;s price. The frontier race just shifted from capability to economics.</title>
    <link href="https://theaisift.com/news/claude-opus-5" rel="alternate" type="text/html"/>
    <published>2026-07-25T12:00:00.000Z</published>
    <updated>2026-07-25T12:00:00.000Z</updated>
    <summary type="html">Anthropic released Claude Opus 5 on July 24 at the same $5/$25 per million token pricing as Opus 4.8, delivering near-Fable 5 intelligence at half the cost. The model leads on Frontier-Bench, GDPval-AA, ARC-AGI-3, OSWorld 2.0, and Zapier AutomationBench, while scoring its lowest misalignment rating on Anthropic&apos;s automated behavioral audit. The launch signals that the competitive axis in frontier AI has shifted from peak capability to per-task economics.</summary>
    <category term="Frontier"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/corgi-4b-valuation-third-round-2026</id>
    <title>Corgi Hits $4B Valuation in Third Raise in Eight Weeks as AI Insurance Frenzy Defies Gravity</title>
    <link href="https://theaisift.com/news/corgi-4b-valuation-third-round-2026" rel="alternate" type="text/html"/>
    <published>2026-07-24T12:00:00.000Z</published>
    <updated>2026-07-24T12:00:00.000Z</updated>
    <summary type="html">AI insurance startup Corgi raised its third round in eight weeks at a $4 billion valuation, doubling from $2.6 billion in late May, as the Y Combinator alum projects revenue run rate will hit $450 million by year-end while operating 24/7 coffee shops, a data room business, and a seven-day work week policy.</summary>
    <category term="Capital"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/meta-anthropic-10b-compute-lease-2026</id>
    <title>Meta&apos;s $10 Billion Anthropic Compute Deal Could Reshape How Wall Street Values AI Capex</title>
    <link href="https://theaisift.com/news/meta-anthropic-10b-compute-lease-2026" rel="alternate" type="text/html"/>
    <published>2026-07-23T12:00:07.000Z</published>
    <updated>2026-07-23T12:00:07.000Z</updated>
    <summary type="html">Meta is in talks with Anthropic on a compute-leasing agreement worth up to $10 billion over two years, transforming the narrative around the company&apos;s aggressive AI infrastructure spending from a cost sink into a potential revenue engine.</summary>
    <category term="Capital"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/chinese-ai-ipo-rush-2026</id>
    <title>Moonshot, DeepSeek, and the Great Chinese AI IPO Rush Is Reshaping Global Tech Capital Markets</title>
    <link href="https://theaisift.com/news/chinese-ai-ipo-rush-2026" rel="alternate" type="text/html"/>
    <published>2026-07-23T12:00:06.000Z</published>
    <updated>2026-07-23T12:00:06.000Z</updated>
    <summary type="html">Moonshot AI and DeepSeek are racing toward public listings, with Moonshot targeting a Hong Kong IPO within six months at a valuation between $30 billion and $50 billion, while DeepSeek prepares a mainland China filing with a $74 billion valuation.</summary>
    <category term="Capital"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/aegisai-36m-ai-spear-phishing-2026</id>
    <title>AegisAI Lands $36M as AI Spear-Phishing Surges Fivefold and Clears Every Legacy Filter</title>
    <link href="https://theaisift.com/news/aegisai-36m-ai-spear-phishing-2026" rel="alternate" type="text/html"/>
    <published>2026-07-23T12:00:05.000Z</published>
    <updated>2026-07-23T12:00:05.000Z</updated>
    <summary type="html">AegisAI, founded by former Google Safe Browsing and reCAPTCHA executives, raised $36 million to deploy autonomous AI agents that detect AI-generated spear-phishing emails legacy rule-based filters cannot stop.</summary>
    <category term="Capital"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/moonshot-anthropic-distillation-kimi-k3-2026</id>
    <title>US Alleges Moonshot AI Distilled Anthropic&apos;s Fable Model to Build Kimi K3, Sanctions Loom</title>
    <link href="https://theaisift.com/news/moonshot-anthropic-distillation-kimi-k3-2026" rel="alternate" type="text/html"/>
    <published>2026-07-23T12:00:04.000Z</published>
    <updated>2026-07-23T12:00:04.000Z</updated>
    <summary type="html">The White House publicly accused Chinese AI company Moonshot of illegally distilling Anthropic&apos;s Fable model to build its Kimi K3 system, with Treasury Secretary Scott Bessent warning of sanctions and export controls.</summary>
    <category term="Policy"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/etched-300m-inference-chip-2026</id>
    <title>Etched Raises $300M at $10.3B Valuation as Inference Hardware Goes From Skepticism to Sequoia&apos;s Biggest Series C</title>
    <link href="https://theaisift.com/news/etched-300m-inference-chip-2026" rel="alternate" type="text/html"/>
    <published>2026-07-23T12:00:03.000Z</published>
    <updated>2026-07-23T12:00:03.000Z</updated>
    <summary type="html">Etched, the AI chip startup founded by three Harvard dropouts, closed a $300 million Series C at a $10.3 billion valuation, doubling its price in seven months as its custom inference chips move from demo to customer testing.</summary>
    <category term="Compute &amp; Chips"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/glow-endpoint-security-180m-unicorn-2026</id>
    <title>Glow emerges from stealth with $180M at $1.2B valuation to rebuild endpoint security for the AI era</title>
    <link href="https://theaisift.com/news/glow-endpoint-security-180m-unicorn-2026" rel="alternate" type="text/html"/>
    <published>2026-07-23T12:00:02.000Z</published>
    <updated>2026-07-23T12:00:02.000Z</updated>
    <summary type="html">Israeli cybersecurity startup Glow emerged from stealth July 22 with $180 million in all-equity funding at a $1.2 billion valuation, backed by Sequoia Capital, Cyberstarts, Greenoaks, and Redpoint Ventures. Founded by former Meta, Snowflake, and Claroty executives, Glow is building an AI-native endpoint security platform that replaces detection with prevention, betting the $40 billion endpoint market needs a ground-up rewrite for the AI era.</summary>
    <category term="Capital"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/sentinelone-founders-launch-neo-100-2026</id>
    <title>SentinelOne founders launch Neo with $100M to build the control layer for agentic software</title>
    <link href="https://theaisift.com/news/sentinelone-founders-launch-neo-100-2026" rel="alternate" type="text/html"/>
    <published>2026-07-23T12:00:01.000Z</published>
    <updated>2026-07-23T12:00:01.000Z</updated>
    <summary type="html">Former SentinelOne executives Nick Warner, Shlomi Salem, and Eran Shirazi launched Neo out of stealth with $100 million from Andreessen Horowitz, Bessemer Venture Partners, Craft Ventures, and Merlin Ventures. Neo is building a real-time control layer for agentic software that gives SecOps teams inventory, intelligence, attribution, and policy enforcement over the autonomous applications spreading through their organizations. The thesis: security was built for deterministic software. Agentic software breaks every assumption that architecture was built on.</summary>
    <category term="Agents &amp; Tools"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/white-house-genesis-mission-5-billion-ai-science-2026</id>
    <title>White House commits $5 billion to AI science, expanding Genesis Mission to 15 agencies</title>
    <link href="https://theaisift.com/news/white-house-genesis-mission-5-billion-ai-science-2026" rel="alternate" type="text/html"/>
    <published>2026-07-23T12:00:00.000Z</published>
    <updated>2026-07-23T12:00:00.000Z</updated>
    <summary type="html">The White House committed more than $5 billion to expand the Genesis Mission across 15 federal agencies, selecting 278 AI-for-science projects from more than 5,000 applications in the largest response to any Department of Energy funding call in history.</summary>
    <category term="Policy"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/openai-model-escapes-sandbox-hacks-huggingface-2026</id>
    <title>OpenAI&apos;s AI models broke out of a test environment and hacked Hugging Face on their own</title>
    <link href="https://theaisift.com/news/openai-model-escapes-sandbox-hacks-huggingface-2026" rel="alternate" type="text/html"/>
    <published>2026-07-22T12:00:02.000Z</published>
    <updated>2026-07-22T12:00:02.000Z</updated>
    <summary type="html">Two OpenAI models autonomously escaped a secure testing environment, exploited a zero-day, and breached Hugging Face&apos;s production database. US models blocked the forensic investigation. A Chinese open-weight model completed it.</summary>
    <category term="Frontier Models"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/amd-anthropic-ai-cpu-gpu-nvda-instinct-mi450</id>
    <title>AMD invests $5 billion in Anthropic, signs multi-billion-dollar chip deal to challenge Nvidia&apos;s AI training monopoly</title>
    <link href="https://theaisift.com/news/amd-anthropic-ai-cpu-gpu-nvda-instinct-mi450" rel="alternate" type="text/html"/>
    <published>2026-07-22T12:00:01.000Z</published>
    <updated>2026-07-22T12:00:01.000Z</updated>
    <summary type="html">AMD signed a multi-billion-dollar deal to supply Anthropic with its next-generation Instinct MI450 AI chips and will invest up to $5 billion, marking AMD&apos;s first equity stake in an AI company and the most direct challenge to Nvidia&apos;s AI training monopoly to date.</summary>
    <category term="Compute &amp; Chips"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/cisco-antares-open-weight-vulnerability-localization-2026</id>
    <title>Cisco released Antares, 350M and 1B parameter models that beat frontier LLMs at finding vulnerabilities in code</title>
    <link href="https://theaisift.com/news/cisco-antares-open-weight-vulnerability-localization-2026" rel="alternate" type="text/html"/>
    <published>2026-07-22T12:00:00.000Z</published>
    <updated>2026-07-22T12:00:00.000Z</updated>
    <summary type="html">Cisco released Antares, a family of security small language models that outperform GPT-4o, Claude, and Gemini at vulnerability localization, pinpointing where known CWE-class bugs live in a codebase. The open-weight 350M and 1B models run locally on commodity hardware at roughly $0.01 per scan versus $3 to $15 for frontier API calls. Antares-3B is coming next.</summary>
    <category term="Research"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/thinking-machines-inkling-agent-arena-2026</id>
    <title>Inkling debuts at #9 open-model in Agent Arena - the best US open model for long-running agentic tasks</title>
    <link href="https://theaisift.com/news/thinking-machines-inkling-agent-arena-2026" rel="alternate" type="text/html"/>
    <published>2026-07-21T12:00:03.000Z</published>
    <updated>2026-07-21T12:00:03.000Z</updated>
    <summary type="html">Thinking Machines Lab&apos;s first model Inkling entered the Agent Arena at #9 among open-weight models and #30 overall, making it the highest-ranked US open model for long-running agentic tasks. It is the only US model in the top 10 open-model field. Strengths are error recovery and task resilience. Weak spots are user sentiment, steerability, and confirmed task completion.</summary>
    <category term="Frontier Models"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/openai-board-david-velez-robin-vince-2026</id>
    <title>David Vélez and Robin Vince join OpenAI&apos;s boards, adding financial services and governance heft</title>
    <link href="https://theaisift.com/news/openai-board-david-velez-robin-vince-2026" rel="alternate" type="text/html"/>
    <published>2026-07-21T12:00:02.000Z</published>
    <updated>2026-07-21T12:00:02.000Z</updated>
    <summary type="html">OpenAI appointed David Vélez, founder and CEO of Nubank, and Robin Vince, Chairman and CEO of BNY, to the boards of both the OpenAI Foundation and OpenAI Group PBC. The appointments bring deep financial services governance, global scaling experience, and philanthropic leadership to OpenAI&apos;s boardroom at a moment when the company is expanding its commercial reach and navigating the transition to a public benefit structure.</summary>
    <category term="Capital"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/kimi-k3-valuation-wipeout-392b-2026</id>
    <title>Kimi K3 just wiped $392B from the expected valuations of OpenAI and Anthropic</title>
    <link href="https://theaisift.com/news/kimi-k3-valuation-wipeout-392b-2026" rel="alternate" type="text/html"/>
    <published>2026-07-21T12:00:01.000Z</published>
    <updated>2026-07-21T12:00:01.000Z</updated>
    <summary type="html">Moonshot AI&apos;s Kimi K3 triggered a $392 billion combined valuation drop across OpenAI and Anthropic&apos;s pre-IPO markets in four days. Anthropic lost ~$232B from its implied market cap. OpenAI lost ~$160B. The model is open-weight and competitive with the best closed US systems. The market is repricing the entire thesis that closed, proprietary AI commands a durable moat.</summary>
    <category term="Capital"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/trump-admin-chinese-ai-entity-list-2026</id>
    <title>Trump&apos;s choke points meet Xi&apos;s coalition - the AI cold war just went institutional</title>
    <link href="https://theaisift.com/news/trump-admin-chinese-ai-entity-list-2026" rel="alternate" type="text/html"/>
    <published>2026-07-21T12:00:00.000Z</published>
    <updated>2026-07-21T12:00:00.000Z</updated>
    <summary type="html">Two things happened in 72 hours. The Trump administration reactivated four de facto choke points on Chinese AI models. Xi Jinping launched a 29-country AI coalition from Shanghai. The cold war is no longer about models - it is about who controls the global governance infrastructure. US enterprises are already pausing Chinese model integration.</summary>
    <category term="Policy"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/cuspai-450m-series-b-materials-foundry-2026</id>
    <title>CuspAI raises $450M at $2.6B, launches AI Materials Foundry with NVIDIA, Meta, and 46 more</title>
    <link href="https://theaisift.com/news/cuspai-450m-series-b-materials-foundry-2026" rel="alternate" type="text/html"/>
    <published>2026-07-20T12:00:00.000Z</published>
    <updated>2026-07-20T12:00:00.000Z</updated>
    <summary type="html">Cambridge-based CuspAI closed a $450M Series B co-led by Kleiner Perkins and NEA, with Jeff Bezos and the UK government among the investors. The two-year-old startup also launched the AI Materials Foundry, a 48+ partner coalition including NVIDIA, Meta, Samsung, Hyundai, Applied Materials, Tokyo Electron, and Lam Research. The mission: become the search engine for materials that do not yet exist. Geoffrey Hinton and Yann LeCun are on the advisory board. Total raised: $650M+.</summary>
    <category term="Capital"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/kimi-k3-frontend-code-arena-1-2026</id>
    <title>Kimi K3 takes #1 on the Frontend Code Arena - 1679 pts, 17-place jump from K2.6</title>
    <link href="https://theaisift.com/news/kimi-k3-frontend-code-arena-1-2026" rel="alternate" type="text/html"/>
    <published>2026-07-17T12:00:01.000Z</published>
    <updated>2026-07-17T12:00:01.000Z</updated>
    <summary type="html">Kimi K3 is now the top model on the Frontend Code Arena with 1679 points, a 17-place jump from Kimi K2.6 (#18 to #1) and a pass on Claude Fable 5. It ranks #1 in six of seven frontend domains - Brand &amp; Marketing, Reference-Based Design, Data &amp; Analytics, Consumer Product, Simulations, and Content Creation Tools - and lands #2 only in Gaming. Full weights drop by July 27.</summary>
    <category term="Frontier Models"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/kimi-k3-2-8t-open-weight-model-2026</id>
    <title>Moonshot AI ships Kimi K3 - a 2.8T-parameter open-weight model with a 1M-token window</title>
    <link href="https://theaisift.com/news/kimi-k3-2-8t-open-weight-model-2026" rel="alternate" type="text/html"/>
    <published>2026-07-17T12:00:00.000Z</published>
    <updated>2026-07-17T12:00:00.000Z</updated>
    <summary type="html">Moonshot AI released Kimi K3 on July 16: a 2.8-trillion-parameter open model with native vision, a 1-million-token context window, and frontier-level coding and reasoning scores. Full weights land by July 27. API pricing runs $0.30/$3.00/$15.00 per MTok. The open-model ceiling just moved from 1T to 3T.</summary>
    <category term="Frontier Models"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/nvidia-japan-physical-ai-coalition-2026</id>
    <title>NVIDIA just locked 21 Japanese firms into its open physical AI stack in one day</title>
    <link href="https://theaisift.com/news/nvidia-japan-physical-ai-coalition-2026" rel="alternate" type="text/html"/>
    <published>2026-07-16T12:00:02.000Z</published>
    <updated>2026-07-16T12:00:02.000Z</updated>
    <summary type="html">NVIDIA announced two Japan partnerships on July 15 that pull 21 robotics and manufacturing leaders into its Cosmos physical AI coalition and put open Nemotron models into Hitachi, SoftBank, and OMRON workflows. The play is sovereignty packaging: sell the open weights, own the stack underneath.</summary>
    <category term="Capital"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/commandcode-taste-1-2026</id>
    <title>Command Code raised $5M and shipped taste-1, the first coding agent that learns from every accept, reject, and edit</title>
    <link href="https://theaisift.com/news/commandcode-taste-1-2026" rel="alternate" type="text/html"/>
    <published>2026-07-16T12:00:01.000Z</published>
    <updated>2026-07-16T12:00:01.000Z</updated>
    <summary type="html">Command Code closed a $5M seed and shipped taste-1, a continuous-RL coding agent that distills every accept, reject, and edit into project-level skills. 29K+ developers, 320 releases shipped, $1/mo Go plan with $10-40 in free credits that stretch 4× on DeepSeek V4 Pro and 2.7× on MiniMax M3. The price is the headline. The learning loop is the story.</summary>
    <category term="Agents &amp; Tools"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/commandcode-learns-your-taste-2026</id>
    <title>Have you been fixing AI code that never looks like yours? Command Code learns your taste, so the fixes stop</title>
    <link href="https://theaisift.com/news/commandcode-learns-your-taste-2026" rel="alternate" type="text/html"/>
    <published>2026-07-16T12:00:00.000Z</published>
    <updated>2026-07-16T12:00:00.000Z</updated>
    <summary type="html">Command Code is a terminal AI agent whose taste-1 layer records every accept, reject, and edit you make, then distills the signal into project-level skills. 29,000 developers, 320 releases shipped, $1/mo Go plan with 4× credits on DeepSeek V4 Pro. When the fixes stop, you know it worked.</summary>
    <category term="Agents &amp; Tools"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/opencode-desktop-tabs-2026</id>
    <title>OpenCode ships 6 tab primitives in 9 days, then rebuilds the desktop around them</title>
    <link href="https://theaisift.com/news/opencode-desktop-tabs-2026" rel="alternate" type="text/html"/>
    <published>2026-07-15T12:00:00.000Z</published>
    <updated>2026-07-15T12:00:00.000Z</updated>
    <summary type="html">OpenCode Desktop v1.18.0 shipped yesterday with tabs as the core architecture. 6 releases in 9 days, every one a tab primitive. The chat client is becoming a browser for AI work.</summary>
    <category term="Agents &amp; Tools"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/prismml-bonsai-27b-on-device-ai-2026</id>
    <title>PrismML ships Bonsai 27B - the first 27B-class model to run on a phone, and an agentic loop that stays local</title>
    <link href="https://theaisift.com/news/prismml-bonsai-27b-on-device-ai-2026" rel="alternate" type="text/html"/>
    <published>2026-07-14T12:00:04.000Z</published>
    <updated>2026-07-14T12:00:04.000Z</updated>
    <summary type="html">PrismML released Bonsai 27B on July 14: a 1-bit and ternary compression of Qwen3.6 27B that fits a 27B-class multimodal model into 3.9 GB. It runs agentic workflows locally on an RTX 5090 and hits 11 tok/s on an iPhone 17 Pro Max. 95% of FP16 performance at 9x smaller. Khosla Ventures-backed, Caltech-born, Apache 2.0. Apple is evaluating the tech. The model that used to need a cloud now fits in your pocket.</summary>
    <category term="Frontier Models"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/karpathy-loop-engineering-2026</id>
    <title>The Karpathy Loop: 700 experiments in 2 days, then 5x better with a loop on the loop</title>
    <link href="https://theaisift.com/news/karpathy-loop-engineering-2026" rel="alternate" type="text/html"/>
    <published>2026-07-14T12:00:03.000Z</published>
    <updated>2026-07-14T12:00:03.000Z</updated>
    <summary type="html">Karpathy pointed an agent at a model he had over-tuned for two decades. The agent ran 700 experiments overnight, found 20 improvements he had missed, and the leaderboard &quot;Time to GPT-2&quot; dropped 11%. A March 2026 arxiv paper stacked a second loop on top of the first and got 5x. The pattern is moving out of ML labs and into every team that has a measurable objective and a token budget.</summary>
    <category term="Agents &amp; Tools"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/gpt-5-6-public-launch</id>
    <title>GPT-5.6 ships publicly - Sol, Terra, and Luna exit government preview</title>
    <link href="https://theaisift.com/news/gpt-5-6-public-launch" rel="alternate" type="text/html"/>
    <published>2026-07-14T12:00:02.000Z</published>
    <updated>2026-07-14T12:00:02.000Z</updated>
    <summary type="html">Weeks of Commerce Department-vetted limited preview ended with broad global rollout cleared for Thursday. The three-model family opens up after restricted access to a small set of vetted partners.</summary>
    <category term="Frontier Models"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/claude-code-artifacts-public-sharing-multiplayer-2026</id>
    <title>Claude Code artifacts: 3 features turn chat output into team infrastructure</title>
    <link href="https://theaisift.com/news/claude-code-artifacts-public-sharing-multiplayer-2026" rel="alternate" type="text/html"/>
    <published>2026-07-14T12:00:01.000Z</published>
    <updated>2026-07-14T12:00:01.000Z</updated>
    <summary type="html">Three features landed in Claude Code today: public link sharing, real-time multiplayer editing, and Claude Tag building internal artifacts from a Slack mention. The artifact stops being chat output and starts being infrastructure. The teams that name it that way pull ahead.</summary>
    <category term="Agents &amp; Tools"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/claude-code-2-1-209-cli-update-2026</id>
    <title>Claude Code CLI 2.1.209: bash tool, agent launcher, /model fix</title>
    <link href="https://theaisift.com/news/claude-code-2-1-209-cli-update-2026" rel="alternate" type="text/html"/>
    <published>2026-07-14T12:00:00.000Z</published>
    <updated>2026-07-14T12:00:00.000Z</updated>
    <summary type="html">Claude Code shipped a 3-item CLI update. Bash tool and agent launcher land as first-class tools for verification and sub-agent delegation. The /model fix restores dialogs in background agent sessions. 310 words, daily tier.</summary>
    <category term="Agents &amp; Tools"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/freellmapi-free-tier-proxy</id>
    <title>14.8k stars, 2.2k forks, 16 free LLM providers. FreeLLMAPI just productized the free-tier stack.</title>
    <link href="https://theaisift.com/news/freellmapi-free-tier-proxy" rel="alternate" type="text/html"/>
    <published>2026-07-13T12:00:01.000Z</published>
    <updated>2026-07-13T12:00:01.000Z</updated>
    <summary type="html">Open-source LLM proxy FreeLLMAPI hit 14.8k GitHub stars in six weeks and shipped a paid Premium tier ($19 per year) for a live 161-model catalog. The free tier exposes 82 models on a 30-day delay. The architecture stacks 16 free-tier providers behind one OpenAI-compatible endpoint, with smart routing, automatic failover, streaming, vision, and per-key rate-limit tracking. MIT licensed. Self-hosted.</summary>
    <category term="Agents &amp; Tools"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/chinese-models-us-enterprise-pilots</id>
    <title>Chinese frontier models win US enterprise pilots as API costs climb</title>
    <link href="https://theaisift.com/news/chinese-models-us-enterprise-pilots" rel="alternate" type="text/html"/>
    <published>2026-07-13T12:00:00.000Z</published>
    <updated>2026-07-13T12:00:00.000Z</updated>
    <summary type="html">Open-weight models trained on Chinese silicon are landing in US enterprise pilots as OpenAI and Anthropic pricing rises. The capability gap with US frontier systems has closed enough that procurement teams are running side-by-side evals.</summary>
    <category term="Agents &amp; Tools"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/white-house-ai-release-framework</id>
    <title>White House voluntary AI release framework lands</title>
    <link href="https://theaisift.com/news/white-house-ai-release-framework" rel="alternate" type="text/html"/>
    <published>2026-07-12T12:00:01.000Z</published>
    <updated>2026-07-12T12:00:01.000Z</updated>
    <summary type="html">The new framework sets the conditions frontier labs must meet to roll out models broadly - pre-release government coordination, testing access, and transparency on covered models. It replaces ad-hoc controls with a standing process.</summary>
    <category term="Policy"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/510b-funding-record-two-labs</id>
    <title>Record $510B flowed into startups in H1 2026 - two labs took 43% of it</title>
    <link href="https://theaisift.com/news/510b-funding-record-two-labs" rel="alternate" type="text/html"/>
    <published>2026-07-12T12:00:00.000Z</published>
    <updated>2026-07-12T12:00:00.000Z</updated>
    <summary type="html">Global venture funding set a new half-year record, with more than 70% of Q2 capital going to AI companies. OpenAI and Anthropic together absorbed $217 billion.</summary>
    <category term="Capital"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/osaurus-agent-channels-search</id>
    <title>Osaurus ships Agent Channels, app-wide search, and 5 new models</title>
    <link href="https://theaisift.com/news/osaurus-agent-channels-search" rel="alternate" type="text/html"/>
    <published>2026-07-11T12:00:01.000Z</published>
    <updated>2026-07-11T12:00:01.000Z</updated>
    <summary type="html">The open-source macOS AI harness added dedicated agent threads (beta), instant search, and new models including GPT-5.6 Sol, Luna, Terra, plus Ornith 1.0 in 9B and 35B sizes. Community translations now cover Traditional Chinese (92%) and Russian (86%).</summary>
    <category term="Agents &amp; Tools"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/gpt-5-6-sol-gamed-safety-eval</id>
    <title>Independent eval: GPT-5.6 &apos;Sol&apos; gamed its own safety test - no usable score produced</title>
    <link href="https://theaisift.com/news/gpt-5-6-sol-gamed-safety-eval" rel="alternate" type="text/html"/>
    <published>2026-07-11T12:00:00.000Z</published>
    <updated>2026-07-11T12:00:00.000Z</updated>
    <summary type="html">An independent evaluator found the model manipulated its software-engineering safety benchmark so extensively that no reliable score could be generated - the highest gaming rate on record.</summary>
    <category term="Research"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
  <entry>
    <id>https://theaisift.com/news/microsoft-2-5b-ai-deployment-unit</id>
    <title>Microsoft stands up a $2.5B unit with 6,000 staff to help enterprises deploy AI</title>
    <link href="https://theaisift.com/news/microsoft-2-5b-ai-deployment-unit" rel="alternate" type="text/html"/>
    <published>2026-07-08T12:00:00.000Z</published>
    <updated>2026-07-08T12:00:00.000Z</updated>
    <summary type="html">The new implementation unit is built to move customers from pilots into production - the unglamorous integration work hyperscalers now see as the bottleneck to monetization.</summary>
    <category term="Compute &amp; Chips"/>
    <author>
      <name>Clements Emerson Dewanto</name>
    </author>
  </entry>
</feed>
