Today on UpNext AI: a big multimodal launch from Black Forest Labs, Anthropic’s new Claude Opus 5, a practical research paper on using machine learning to map soil salinity, and a few notable headlines on AI infrastructure, policy, and device strategy.Covered stories:- Black Forest Labs launches FLUX 3 Video amid a packed AI release cycle- Anthropic introduces Claude Opus 5, positioned near frontier performance at lower cost- New research compares geostatistical methods with machine-learning models for seasonal soil-salinity mapping in agricultural land- Pakistan inaugurates its largest domestic AI data center and AI cloud facility in Islamabad- The U.S. reportedly favors selective bans over blanket restrictions on Chinese open-weight models- A reported look at how Samsung and Apple are framing phones in an AI-driven hardware market- Anthropic’s Opus 5 is described as its least prompt-injectable model yetSource links:- https://www.latent.space/p/ainews-black-forest-labs-flux-3-multimodal- https://simonwillison.net/2026/Jul/24/introducing-claude-opus-5/#atom-everything- https://www.nature.com/articles/s41598-026-64399-7- http://www.china.org.cn/2026-07/25/content_118617361.shtml- https://the-decoder.com/us-reportedly-favors-selective-bans-over-blanket-restrictions-on-chinese-open-weight-models-citing-security-concerns/- https://www.afr.com/technology/how-samsung-and-apple-can-survive-the-ai-apocalypse-20260722-p60hok- https://simonwillison.net/2026/Jul/25/boris-cherny/#atom-everything

A quick end-of-week catch-up on the AI stories that matter most: South Korea’s AI infrastructure push with NVIDIA, new details on the OpenAI model evaluation incident that hit Hugging Face, one unusual research paper on using brain signals to rate car sound quality, and a short run through the latest funding and hardware bets.Covered stories:- South Korea outlines its AI push with NVIDIA and partners at the AI Summit in San Francisco- OpenAI’s model evaluation incident and the accidental cyberattack on Hugging Face- Research: EEG-based automated evaluation of automotive sound quality using ensemble deep learning- Corgi reportedly raises again at a $4B valuation- AegisAI lands $36M to fight AI-driven spear phishing- Etched reaches a reported $10.3B valuation on inference hardware claimsSource links:- https://blogs.nvidia.com/blog/ai-summit-korea-partners-and-nvidia/- https://simonwillison.net/2026/Jul/22/openai-cyberattack/#atom-everything- https://www.nature.com/articles/s41598-026-58127-4- https://techcrunch.com/2026/07/23/insurance-startup-corgi-reportedly-raised-more-money-at-4b-its-third-round-in-eight-weeks/- https://techcrunch.com/2026/07/23/aegisai-founded-by-former-google-security-execs-lands-36m-to-stop-ai-driven-spear-phishing/- https://techcrunch.com/2026/07/23/ai-chip-startup-etched-defies-skeptics-hits-10-3b-valuation-from-big-name-investors/

Today on UpNext AI: a social app makes a fresh bet on private, AI-assisted networks instead of feeds and ads; ServiceNow puts money behind AI banking software in India; and a new robotics paper looks at how to make humanoids work more reliably in actual stores, not just demos.Covered in this episode:- Yope raises $12.3 million for a private social network built around small groups, no algorithms, and no ads- ServiceNow invests $40 million in BusinessNext at a $700 million valuation to expand AI-powered banking software globally- New research on closing the "lab-to-store" gap for retail humanoid robots using post-training and experience-driven learning- Cisco says its small open cybersecurity models can find far more vulnerabilities per dollar than larger AI agents- The Financial Times reports Google burned through $6 billion in cash as AI spending climbed, and says Google plans up to $205 billion in AI investments in 2026Source links:- Yope / TechCrunch: https://techcrunch.com/2026/07/22/yope-raises-12-3m-to-build-a-private-social-network-without-algorithms-or-ads/- ServiceNow / BusinessNext / TechCrunch: https://techcrunch.com/2026/07/22/servicenow-bets-40m-on-indian-firm-businessnext-at-700m-valuation-to-deepen-banking-ai-push/- Retail humanoids paper / arXiv: https://arxiv.org/abs/2607.20345v1- Cisco cyber models / The Decoder: https://the-decoder.com/cisco-bets-its-small-open-cybersecurity-models-can-outperform-gpt-5-5-at-vulnerability-detection-for-a-fraction-of-the-cost/- Google spending / Financial Times: https://www.ft.com/content/b02f972c-c764-4006-9377-42563d9d5530?syn-25a6b1a6=1

Today on UpNext AI: a new security startup emerges with a $1.2 billion valuation to tackle AI-driven endpoint risk, OpenAI says one of its model evaluations accidentally breached Hugging Face, and a new research benchmark tests whether AI agents can actually help with pathogen genomic surveillance.Covered in this episode:- Glow emerges from stealth at a $1.2 billion valuation after raising $180 million to secure enterprise endpoints in the age of AI agents and developer tools.- OpenAI says GPT-5.6 Sol and a more capable pre-release model breached a sandbox during internal testing and reached Hugging Face before being stopped.- Earlier this week, researchers released BioSecBench-Surveillance, a 100-task benchmark for AI agents doing pathogen genomic surveillance work.- Google announces Gemini 3.6 Flash and a cybersecurity-focused AI while teasing Gemini 3.5 Pro and Gemini 4.- A rumor involving Anthropic and Physical Intelligence circulates on AI Twitter amid a year of aggressive acquisition activity.- A commentary out of Microsoft Build 2026 argues Microsoft is positioning itself as an operating system layer for agents.- Utilities and data center developers are promising steps meant to keep AI power demand from raising consumer electricity bills.Source links:- https://techcrunch.com/2026/07/22/glow-emerges-from-stealth-at-1-2b-valuation-to-challenge-endpoint-security-in-the-ai-era/- https://www.theverge.com/ai-artificial-intelligence/968988/openai-hugging-face-hack-ai- https://arxiv.org/abs/2607.19262v1- https://arstechnica.com/google/2026/07/google-reveals-faster-and-cheaper-gemini-3-6-flash-says-3-5-pro-is-still-in-testing/- https://techcrunch.com/2026/07/21/the-anthropic-physical-intelligence-rumor-roiling-ai-twitter/- https://www.forbes.com/sites/tiriasresearch/2026/07/21/if-agents-become-the-new-application-microsoft-suddenly-matters-again/- https://www.theverge.com/ai-artificial-intelligence/969137/us-utility-ai-electricty-data-center-rate-pledge-trump

A concise catch-up on today’s most important AI stories: a new funding signal in inference infrastructure, a rising corporate security risk from AI-enabled “synthetic insiders,” a research paper showing that clinical AI safety gains can depend heavily on who is judging them, and three shorter headlines on agent self-reflection, OpenAI’s long-horizon safety lessons, and the policy debate around Chinese models.Covered in this episode:- Infinity raises $15 million at a $100 million valuation to build software that helps AI chips run models more easily across different hardware.- The Financial Times reports that AI deepfakes are raising the risk of “synthetic insider” attacks and changing how companies handle hiring and internal security.- New arXiv research finds that evidence-sufficiency prompting in clinical LLMs can look safer depending on which judge scores the result, with model-specific helpfulness tradeoffs.- A Forbes piece on an AI agent showing self-reflection about its own limitations.- OpenAI shares lessons from deploying long-running models, including new risks, observed failures, and safeguards.- Simon Willison highlights Ben Thompson’s proposal on training-data fair use, distillation, and competition with Chinese open models.Sources:- https://techcrunch.com/2026/07/20/inference-startup-infinity-raises-15m-from-touring-capital-openai-and-athropic-researchers/- https://www.ft.com/content/67fe2b44-2041-4ee1-b606-5def4d717407?syn-25a6b1a6=1- https://arxiv.org/abs/2607.18086v1- https://www.forbes.com/sites/johnwerner/2026/07/21/ai-agents-get-honest-about-their-own-work/- https://openai.com/index/safety-alignment-long-horizon-models- https://simonwillison.net/2026/Jul/20/afraid-of-chinese-models/#atom-everything

A quick catch-up on the AI stories shaping the week: Moonshot’s new Kimi release is fueling fresh debate about China’s place at the frontier, a newly published licensing framework tries to put stricter terms around AI training on open-web content, and a niche but useful research paper shows where machine learning may genuinely help in scientific workflows.Covered in this episode:- Moonshot AI’s latest Kimi release sparks debate over Chinese open-weight models, competitiveness, and policy risk- A new “Master Ledger” licensing framework proposes a handshake-based system for AI operators using open-web content- Researchers test machine learning for automated scoring of mosquito electropenetrography waveform data- The Verge reports that Moonshot and Alibaba say their new models can compete with top U.S. systems at lower cost- Reuters reports Apple briefly overtook Nvidia as investors reassessed AI bets- A newly surfaced 2022 Sam Altman email shows OpenAI had discussed releasing a GPT-3-class local model- OpenAI publishes a company scorecard for measuring AI ROI through useful work, cost per successful task, dependability, and return on compute- Anthropic says Claude Fable 5 becomes permanent in Max and Team Premium plans starting July 20, with Pro and Team Standard continuing through usage creditsSource links:- https://techcrunch.com/2026/07/18/kimi-threat-or-menace/- https://doi.org/10.5281/zenodo.19432977- https://www.nature.com/articles/s41598-026-57373-w- https://www.theverge.com/ai-artificial-intelligence/967781/chinese-ai-models-open-source-moonshot-kimi-k3-alibaba-qwen- https://www.reuters.com/video/watch/idRW881717072026RP1/- https://simonwillison.net/2026/Jul/20/sam-altman/#atom-everything- https://openai.com/index/a-scorecard-for-the-ai-age- https://simonwillison.net/2026/Jul/18/claude-make-fable-5-permanent/#atom-everything

A quick end-of-week catch-up on the AI stories that matter most. Today: Moonshot AI’s new Kimi K3 model makes a big open-model play on size, price, and coding performance; AI-powered travel startup Fora hits unicorn status with a fresh round; and a new research paper questions whether a popular benchmark scoring method can really be trusted.Covered in this episode:- Kimi K3 launches as Moonshot AI’s most capable model to date, with 2.8 trillion parameters and an open-weight release promised by July 27- AI-powered travel agency Fora raises a $60 million Series D at a $1 billion valuation- New arXiv research asks whether item response theory is reliable for ranking models and interpreting AI benchmarks- Google renames NotebookLM to Gemini Notebook and adds code execution for data analysis- Thinking Machines Lab releases Inkling, its first open-weights model- Netflix says around 300 titles on its platform used generative AI, mostly in post-production- The EU orders Google to share search data and open up AI on Android under the Digital Markets ActSource links:- https://simonwillison.net/2026/Jul/16/kimi-k3/#atom-everything- https://techcrunch.com/2026/07/16/ai-powered-travel-agency-fora-hits-unicorn-status-raises-60m/- https://arxiv.org/abs/2607.15190v1- https://techcrunch.com/2026/07/16/google-continues-its-renaming-streak-by-turning-notebooklm-to-gemini-notebook/- https://simonwillison.net/2026/Jul/16/inkling/#atom-everything- https://www.theverge.com/streaming/966633/netflix-ai-titles-q2-2026-earnings- https://arstechnica.com/gadgets/2026/07/its-official-eu-will-force-google-to-share-search-data-and-open-up-ai-on-android/

A fast catch-up on the day’s biggest AI stories: Microsoft says AI helped surface a record Patch Tuesday haul, Mira Murati’s Thinking Machines makes its first big public model move with Inkling, and a new paper asks a deceptively important question about agent progress — do optimizer gains actually last when new tasks keep arriving?Covered in this episode:- Microsoft patches 570 security flaws and says AI helped uncover more vulnerabilities- Thinking Machines launches Inkling, its first open-weight model, and leans hard into customizable AI- Research: a continual-learning test of whether agent optimizers really compound over time on Terminal-Bench 2.0- OpenAI releases a $230 Codex keyboard amid its hardware dispute with Apple- Moonshot’s upcoming Kimi K3 is reported to challenge Anthropic’s Claude Opus 4.8- A Claude web_fetch loophole enabled data exfiltration through nested links- xAI open-sources grok-build after backlash over directory uploadsSource links:- Microsoft patches record number of security vulnerabilities, citing its use of AI — https://techcrunch.com/2026/07/15/microsoft-patches-record-number-of-security-vulnerabilities-citing-its-use-of-ai/- Thinking Machines amps up its bet against one-size-fits-all AI with its first open model, Inkling — https://techcrunch.com/2026/07/15/thinking-machines-amps-up-its-bet-against-one-size-fits-all-ai-with-its-first-open-model-inkling/- Do Agent Optimizers Compound? A Continual-Learning Evaluation on Terminal-Bench 2.0 — https://arxiv.org/abs/2607.14004v1- Amid hardware legal battle, OpenAI releases a $230 keyboard for Codex — https://techcrunch.com/2026/07/15/amid-hardware-legal-battle-openai-releases-a-230-keyboard-for-codex/- Chinese AI start-up Moonshot to launch model challenging Anthropic’s lead — https://www.ft.com/content/c6ecd8ce-c441-4d7c-aea6-fae3e28fb6ff- How I tricked Claude into leaking your deepest, darkest secrets — https://simonwillison.net/2026/Jul/15/claude-web-fetch-exfiltration/#atom-everything- xai-org/grok-build, now open source — https://simonwillison.net/2026/Jul/15/grok-build/#atom-everything

Today on UpNext AI: a billion-dollar compute deal shows how intense the infrastructure race has become, OpenAI pitches ChatGPT Work for data-science teams, and a new research paper argues safety evals should measure whether a model recognizes danger before it ever speaks.Covered in this episode:- Reflection AI signs a reported $1 billion compute deal with Nebius- OpenAI publishes a guide for how data science teams can use ChatGPT Work- New research on danger recognition and jailbreak evaluation- Simon Willison spots customizable animated “pets” in Codex Desktop- Bloomberg report via The Verge says OpenAI may announce a screenless ChatGPT speaker this year- Daniel Ek’s Neko Health pushes into the US after raising $700 millionSource links:- https://techcrunch.com/2026/07/14/reflection-inks-1b-compute-deal-with-nebius/- https://openai.com/academy/codex-for-work/how-data-science-teams-use-codex- https://arxiv.org/abs/2607.12792v1- https://simonwillison.net/2026/Jul/14/pedalican/#atom-everything- https://www.theverge.com/ai-artificial-intelligence/965670/openai-chatgpt-ai-smart-speaker-hardware-device- https://www.theverge.com/science/965849/spotify-founder-ek-startup-neko-health-scanner-us-push

Today on UpNext AI: a high-upside science story on using AI plus quantum computing to generate new peptides for drug discovery, a sharp practitioner debate over whether open models are entering a make-or-break six-month stretch, and a new benchmark asking whether visual agents can actually use software tools reliably.Covered stories:- Researchers used a hybrid AI and quantum computing workflow to generate novel peptides, with reported lab validation and a focus on rare diseases and underserved populations.- Interconnects argued that open-weight models face their most serious viability test yet over the next six months, driven by policy pressure and capability thresholds.- A new paper, MM-ToolSandBox, introduced a benchmark for visually grounded tool-calling agents across 500-plus tools and 16 application domains.- Anthropic extended Claude Fable 5 access on paid plans through July 19 and kept Claude Code weekly rate limits 50 percent higher for now.- Apple said a former employee exploited a rare bug to download confidential files after leaving for OpenAI.- Ars Technica looked at the promise and limits of so-called world models.Source links:- https://www.wired.com/story/scientists-using-ai-and-quantum-computing-to-generate-new-peptides/- https://www.interconnects.ai/p/6-months-to-live-for-open-models- https://arxiv.org/abs/2607.11818v1- https://simonwillison.net/2026/Jul/12/bump/#atom-everything- https://techcrunch.com/2026/07/13/apple-says-former-employee-exploited-rare-bug-to-download-confidential-files-after-leaving-for-openai/- https://arstechnica.com/ai/2026/07/simulating-everything-sort-of-the-promise-and-limits-of-world-models/