# Ground Truth

> AI, checked against the source. Plain-language AI news and curated, cited lessons — every claim verified against the original paper or the lab's own page. No aggregator hearsay, no AI slop.

Built for your AI too: news and lessons are also published as a structured feed (/feed.json) and an LLM front-door (/llms.txt), so your own model or agent can read it.

Full text of every article and lesson, inlined for ingestion: /llms-full.txt

## News (verified, source-linked)
- [The US government just banned Anthropic's most powerful AI model](https://groundtruth.day/news/the-us-government-banned-anthropics-most-powerful-ai-model.html) (2026-06-25) — For the first time, Washington has export-controlled an AI model itself, not the chips it runs on. Anthropic's Fable 5 and Mythos 5 have been dark worldwide since June 12, and the trigger involved an NSA test that the internet has badly misread. · source: https://www.anthropic.com/news/fable-mythos-access · data: https://groundtruth.day/news/the-us-government-banned-anthropics-most-powerful-ai-model.json
- [OpenAI designs its own chip to run its models](https://groundtruth.day/news/openai-designs-its-own-chip-to-run-its-models.html) (2026-06-25) — With Broadcom, OpenAI unveiled a custom chip built for one job: serving its AI models cheaply. · source: https://arstechnica.com/gadgets/2026/06/openai-and-broadcom-announce-chip-designed-for-llm-inference-at-scale/ · data: https://groundtruth.day/news/openai-designs-its-own-chip-to-run-its-models.json
- [Qualcomm buys the software that lets AI run anywhere](https://groundtruth.day/news/qualcomm-buys-modular-and-the-mind-behind-it.html) (2026-06-25) — Qualcomm is paying about $3.9 billion for Modular, the Mojo language, and legendary compiler engineer Chris Lattner. · source: https://www.modular.com/blog/qualcomm-to-acquire-modular · data: https://groundtruth.day/news/qualcomm-buys-modular-and-the-mind-behind-it.json
- [Google's fast model can now use a computer by itself](https://groundtruth.day/news/geminis-fast-model-can-now-use-a-computer.html) (2026-06-25) — Gemini 3.5 Flash gained built-in 'computer use,' letting one model click, type, and act across browsers, phones, and desktops. · source: https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-computer-use-gemini-3-5-flash/ · data: https://groundtruth.day/news/geminis-fast-model-can-now-use-a-computer.json
- [A language model that doesn't write left to right](https://groundtruth.day/news/a-language-model-that-doesnt-write-left-to-right.html) (2026-06-25) — iLLaDA is an 8-billion-parameter model that generates text by refining a blurry whole rather than one word at a time, and it's catching up to the mainstream. · source: https://arxiv.org/abs/2606.25331 · data: https://groundtruth.day/news/a-language-model-that-doesnt-write-left-to-right.json
- [One model that listens, sees, and talks back in real time](https://groundtruth.day/news/one-model-that-listens-sees-and-talks-back-live.html) (2026-06-25) — Wan-Streamer collapses the usual chain of separate speech and video tools into a single model built for live, two-way conversation. · source: https://huggingface.co/papers/2606.25041 · data: https://groundtruth.day/news/one-model-that-listens-sees-and-talks-back-live.json
- [NVIDIA shrinks video generation down to real time](https://groundtruth.day/news/nvidia-shrinks-video-ai-down-to-real-time.html) (2026-06-25) — A new NVIDIA recipe distills slow video-generating AI into a fast version that can stream frames live and react to your actions. · source: https://arxiv.org/abs/2606.25473 · data: https://groundtruth.day/news/nvidia-shrinks-video-ai-down-to-real-time.json
- [A safety switch an AI agent can't reach](https://groundtruth.day/news/a-safety-switch-an-ai-agent-cant-reach.html) (2026-06-25) — Researchers propose putting an agent's safety controls outside the agent itself, so a misbehaving AI structurally cannot turn them off. · source: https://arxiv.org/abs/2606.26057 · data: https://groundtruth.day/news/a-safety-switch-an-ai-agent-cant-reach.json
- [What does your AI actually remember about you?](https://groundtruth.day/news/what-does-your-ai-actually-remember-about-you.html) (2026-06-25) — Two new studies stop trusting that agent 'memory' works and start measuring it directly, with results that carry a privacy sting. · source: https://arxiv.org/abs/2606.24595 · data: https://groundtruth.day/news/what-does-your-ai-actually-remember-about-you.json
- [When AI safety training withholds what could help you](https://groundtruth.day/news/when-ai-safety-training-withholds-what-could-help-you.html) (2026-06-25) — A pre-registered study finds heavily safety-trained models give doctors medical information they refuse to give ordinary people, with identical facts. · source: https://arxiv.org/abs/2604.07709 · data: https://groundtruth.day/news/when-ai-safety-training-withholds-what-could-help-you.json
- [Are closed AI models overpriced luxury goods?](https://groundtruth.day/news/are-closed-ai-models-overpriced-luxury-goods.html) (2026-06-25) — An essay argues open-weight models now undercut the big closed AIs by huge margins, and that 'China fears' are being used to protect those prices. · source: https://jamesoclaire.com/2026/06/25/the-unbearable-cheapness-of-open-weight-models/ · data: https://groundtruth.day/news/are-closed-ai-models-overpriced-luxury-goods.json
- [NVIDIA's warm-water fix for AI's thirsty data centers](https://groundtruth.day/news/nvidias-warm-water-fix-for-ai-thirsty-data-centers.html) (2026-06-25) — A new NVIDIA cooling design claims to use almost no water inside the data center, though critics say that's only part of AI's water bill. · source: https://blogs.nvidia.com/blog/liquid-cooling-ai-factories/ · data: https://groundtruth.day/news/nvidias-warm-water-fix-for-ai-thirsty-data-centers.json
- [A senator says a banned AI broke into nearly all NSA systems in hours](https://groundtruth.day/news/mythos-broke-into-nearly-all-nsa-systems-in-hours.html) (2026-06-24) — New testimony reframes the Mythos export ban: a top general reportedly told a senator the model breached almost all classified systems in a red-team test, not in weeks but in hours. · source: https://securityaffairs.com/194016/ai/anthropics-mythos-ai-broke-into-almost-all-nsa-classified-systems-in-hours.html · data: https://groundtruth.day/news/mythos-broke-into-nearly-all-nsa-systems-in-hours.json
- [Alibaba's new models let AI agents practice in a world they imagine](https://groundtruth.day/news/qwen-agentworld-agents-that-simulate-their-own-world.html) (2026-06-24) — Qwen-AgentWorld trains a model to simulate the environment an agent acts in, then uses that simulation as a cheap, controllable place to learn -- reporting gains beyond training in the real thing. · source: https://arxiv.org/abs/2606.24597 · data: https://groundtruth.day/news/qwen-agentworld-agents-that-simulate-their-own-world.json
- [This model's job is to make better training data for other models](https://groundtruth.day/news/dataclaw0-an-agent-that-prepares-its-own-training-data.html) (2026-06-24) — DataClaw0 turns the grind of cleaning and labeling training data into a learned skill -- a small model that refines raw, messy multimodal streams into dense, purpose-built lessons. · source: https://arxiv.org/abs/2606.21337 · data: https://groundtruth.day/news/dataclaw0-an-agent-that-prepares-its-own-training-data.json
- [An open project publishes the recipe for training capable AI agents](https://groundtruth.day/news/openthoughts-agent-open-recipes-for-training-agents.html) (2026-06-24) — OpenThoughts-Agent releases its full data-curation pipeline, dataset, and experiments -- showing that what an agent learns from matters more than raw size, and letting anyone reproduce it. · source: https://arxiv.org/abs/2606.24855 · data: https://groundtruth.day/news/openthoughts-agent-open-recipes-for-training-agents.json
- [Uber reportedly burned through its whole 2026 AI coding budget in four months](https://groundtruth.day/news/uber-burned-its-ai-budget-in-four-months.html) (2026-06-24) — The clearest enterprise cost figure yet for AI coding agents: Uber's CTO is reported to have said the company exhausted its Claude Code budget in a third of the year. · source: https://www.forbes.com/sites/janakirammsv/2026/05/17/uber-burns-its-2026-ai-budget-in-four-months-on-claude-code/ · data: https://groundtruth.day/news/uber-burned-its-ai-budget-in-four-months.json
- [A small but elegant idea: putting 'experts' inside the attention layer](https://groundtruth.day/news/grouped-query-experts-moe-moves-into-attention.html) (2026-06-24) — Grouped Query Experts brings the mixture-of-experts trick into attention, activating only half a model's query heads per token while matching the full version -- at least at small scale. · source: https://arxiv.org/abs/2606.20945 · data: https://groundtruth.day/news/grouped-query-experts-moe-moves-into-attention.json
- [Anthropic gives AI agents their own work accounts, not yours](https://groundtruth.day/news/claude-agents-get-their-own-identity-at-work.html) (2026-06-24) — Anthropic's new 'agent identity' model lets Claude agents hold their own scoped accounts for tools like GitHub and Slack, tied to channels -- instead of borrowing a human employee's login. · source: https://www.claude.com/blog/agent-identity-access-model · data: https://groundtruth.day/news/claude-agents-get-their-own-identity-at-work.json
- [Can an AI agent match real published science? A new test says: rarely](https://groundtruth.day/news/naturebench-can-coding-agents-do-real-science.html) (2026-06-24) — NatureBench pits coding agents against the published state-of-the-art from Nature-family papers. Even the best agents beat the bar on a small minority of tasks -- mostly by reframing, not inventing. · source: https://arxiv.org/abs/2606.24530 · data: https://groundtruth.day/news/naturebench-can-coding-agents-do-real-science.json
- [Google promised Gemini 3.5 Pro in June. June is almost over.](https://groundtruth.day/news/gemini-3-5-pro-is-running-late.html) (2026-06-24) — Google said its next flagship would arrive in June; with days left it's still limited preview. The timing is awkward -- it overlaps a gap where another Western flagship is also unavailable. · source: https://blog.google/technology/google-deepmind/ · data: https://groundtruth.day/news/gemini-3-5-pro-is-running-late.json
- [An AI Reportedly Broke Into Nearly All of the NSA's Classified Systems in Hours](https://groundtruth.day/news/an-ai-broke-into-nearly-all-the-nsas-classified-systems-in-hours.html) (2026-06-24) — A senator says the head of the NSA told him a top AI model walked through almost all of America's classified systems in hours during a controlled test, reframing last week's government shutdown of the model. · source: https://securityaffairs.com/194016/ai/anthropics-mythos-ai-broke-into-almost-all-nsa-classified-systems-in-hours.html · data: https://groundtruth.day/news/an-ai-broke-into-nearly-all-the-nsas-classified-systems-in-hours.json
- [AI Agents Are Learning to Build the Worlds They Train In](https://groundtruth.day/news/ai-agents-are-learning-to-build-the-worlds-they-train-in.html) (2026-06-24) — Three new open research projects point the same way: instead of only learning what to do, agents are learning to simulate the environment itself, so they can practice in their own imagination. · source: https://arxiv.org/abs/2606.24597 · data: https://groundtruth.day/news/ai-agents-are-learning-to-build-the-worlds-they-train-in.json
- [Microsoft's CEO Says the AI Industry Has Not Earned the Right to Do This](https://groundtruth.day/news/microsofts-ceo-says-the-ai-industry-has-not-earned-the-right.html) (2026-06-24) — In a Wall Street Journal interview, Satya Nadella named OpenAI and Anthropic -- two companies Microsoft has poured billions into -- and warned that an economy reshaped by a handful of AI models will not survive politically. · source: https://www.techtimes.com/articles/318809/20260621/nadella-names-openai-anthropic-ai-giants-must-earn-societal-permission.htm · data: https://groundtruth.day/news/microsofts-ceo-says-the-ai-industry-has-not-earned-the-right.json
- [A Coding AI Ran Through Uber's Yearly Budget in Four Months](https://groundtruth.day/news/a-coding-ai-ran-through-ubers-yearly-budget-in-four-months.html) (2026-06-24) — Uber gave Claude Code to about 5,000 engineers, who loved it. By April the company had burned through its entire 2026 AI budget, exposing how badly old software pricing fits new agent tools. · source: https://www.forbes.com/sites/janakirammsv/2026/05/17/uber-burns-its-2026-ai-budget-in-four-months-on-claude-code/ · data: https://groundtruth.day/news/a-coding-ai-ran-through-ubers-yearly-budget-in-four-months.json
- [A Classic Efficiency Trick Just Moved Into a New Part of the AI](https://groundtruth.day/news/a-classic-efficiency-trick-just-moved-into-a-new-part-of-the-ai.html) (2026-06-24) — For years, the committee-of-specialists design that keeps big models fast lived in one layer of the network. A clean new result shows it works in the attention layer too, halving some of the work for free. · source: https://arxiv.org/abs/2606.20945 · data: https://groundtruth.day/news/a-classic-efficiency-trick-just-moved-into-a-new-part-of-the-ai.json
- [Can an AI Agent Reproduce Real Science? A New Test Says: Rarely](https://groundtruth.day/news/can-an-ai-agent-reproduce-real-science-a-new-test-says-rarely.html) (2026-06-24) — A new benchmark points coding agents at the actual computational results behind ninety papers in top journals. The strongest models matched the published science on fewer than one in five. · source: https://arxiv.org/abs/2606.24530 · data: https://groundtruth.day/news/can-an-ai-agent-reproduce-real-science-a-new-test-says-rarely.json
- [Anthropic Gives Its AI Agents Their Own Logins, Not Yours](https://groundtruth.day/news/anthropic-gives-its-ai-agents-their-own-logins-not-yours.html) (2026-06-24) — As AI agents start working in teams alongside people, the old 'the bot acts as you' model breaks down. Anthropic's answer: give each agent its own scoped account in every system it touches. · source: https://claude.com/blog/agent-identity-access-model · data: https://groundtruth.day/news/anthropic-gives-its-ai-agents-their-own-logins-not-yours.json
- [The Model Ban Is Quietly Redrawing the AI Map](https://groundtruth.day/news/the-model-ban-is-quietly-redrawing-the-ai-map.html) (2026-06-24) — Two weeks after the US pulled its top models off the market, a Chinese open model sits atop the global download charts and the community is busy rebuilding the banned capability in the open. · source: https://huggingface.co/zai-org/GLM-5.2 · data: https://groundtruth.day/news/the-model-ban-is-quietly-redrawing-the-ai-map.json
- [DeepMind Sketches Four Roads From Human-Level AI to Superintelligence](https://groundtruth.day/news/deepmind-sketches-four-roads-from-human-level-ai-to-superintelligence.html) (2026-06-24) — A new report from senior DeepMind researchers lays out four ways AI could push past human-level ability -- and argues the leap is more likely to be a steady climb than a single dramatic jump. · source: https://arxiv.org/abs/2606.12683 · data: https://groundtruth.day/news/deepmind-sketches-four-roads-from-human-level-ai-to-superintelligence.json
- [Samsung Banned ChatGPT in 2023. Now It's Giving It to 125,000 Workers.](https://groundtruth.day/news/samsung-banned-chatgpt-in-2023-now-its-giving-it-to-125000-workers.html) (2026-06-24) — After barring ChatGPT over a data leak three years ago, Samsung has reversed course and rolled OpenAI's enterprise tools out across its workforce -- a vivid sign that the corporate holdouts are capitulating. · source: https://www.pymnts.com/artificial-intelligence/2026/06/samsung-rolls-out-openai-tools-to-workforce/ · data: https://groundtruth.day/news/samsung-banned-chatgpt-in-2023-now-its-giving-it-to-125000-workers.json
- [Sometimes the AI Knew the Better Answer a Few Layers Early](https://groundtruth.day/news/sometimes-the-ai-knew-the-better-answer-a-few-layers-early.html) (2026-06-24) — A new paper finds that a model's final layer can actually muddy an answer its middle layers had right -- and that reading the answer out a little early can claw back ability lost to safety training. · source: https://arxiv.org/abs/2606.21906 · data: https://groundtruth.day/news/sometimes-the-ai-knew-the-better-answer-a-few-layers-early.json
- [The AI That Now Writes Most of Its Maker's Code](https://groundtruth.day/news/claude-now-writes-most-of-anthropics-own-code.html) (2026-06-23) — Anthropic says more than 80 percent of the code it ships is now written by its own model, Claude, and the more interesting numbers are about judgment. · source: https://www.anthropic.com/institute/recursive-self-improvement · data: https://groundtruth.day/news/claude-now-writes-most-of-anthropics-own-code.json
- [Anthropic Wants a Pause Button the Whole World Can Check](https://groundtruth.day/news/anthropic-wants-a-pause-button-the-world-can-check.html) (2026-06-23) — Buried in Anthropic's essay is a concrete proposal: not to stop AI, but to build the machinery that would let rival labs prove to each other they had stopped. · source: https://www.anthropic.com/institute/recursive-self-improvement · data: https://groundtruth.day/news/anthropic-wants-a-pause-button-the-world-can-check.json
- [A Free Model That Splits Your Work Across 300 Helpers](https://groundtruth.day/news/kimi-k2-6-open-model-runs-300-agents-at-once.html) (2026-06-23) — Moonshot AI's Kimi K2.6 is a frontier-grade model anyone can download, and its headline trick is fanning a single job out to hundreds of helpers working in parallel. · source: https://huggingface.co/moonshotai/Kimi-K2.6 · data: https://groundtruth.day/news/kimi-k2-6-open-model-runs-300-agents-at-once.json
- [The US government made a top AI model disappear three days after launch](https://groundtruth.day/news/the-government-pulled-a-frontier-model.html) (2026-06-22) — Washington forced Anthropic to switch off its two most powerful new models worldwide, turning AI export control into something that can happen overnight. · source: https://www.anthropic.com/news · data: https://groundtruth.day/news/the-government-pulled-a-frontier-model.json
- [An AI wrote a working operating-system kernel from scratch in 38 minutes](https://groundtruth.day/news/the-model-that-wrote-a-kernel-in-38-minutes.html) (2026-06-22) — A blow-by-blow log shows one of the now-suspended models building bootable low-level systems code from an empty folder -- the kind of feat that made regulators nervous. · source: https://tolmo.com/blog/when-the-model-writes-the-kernel/ · data: https://groundtruth.day/news/the-model-that-wrote-a-kernel-in-38-minutes.json
- [OpenAI launches a security push at the exact moment its rival got banned](https://groundtruth.day/news/openai-pitches-itself-as-the-safe-cyber-lab.html) (2026-06-22) — Daybreak and 'Patch the Planet' position OpenAI as the responsible cyber-AI lab -- a defensive-security launch whose timing is the whole message. · source: https://openai.com/index/patch-the-planet/ · data: https://groundtruth.day/news/openai-pitches-itself-as-the-safe-cyber-lab.json
- [Suddenly, downloadable AI models look like an insurance policy](https://groundtruth.day/news/open-weights-become-an-insurance-policy.html) (2026-06-22) — With a top hosted model pulled overnight, a flood of powerful open models you can run yourself -- and run fast -- is being reframed from hobby to risk management. · source: https://artificialanalysis.ai/articles/aa-briefcase · data: https://groundtruth.day/news/open-weights-become-an-insurance-policy.json
- [Sakana's new model isn't a model -- it's a committee of models behind one door](https://groundtruth.day/news/one-model-that-is-really-a-committee.html) (2026-06-22) — Fugu routes each request across several frontier AIs and answers through a single endpoint, pitched explicitly as a hedge against depending on any one provider. · source: https://sakana.ai/fugu/ · data: https://groundtruth.day/news/one-model-that-is-really-a-committee.json
- [Two labs race to make AI write whole paragraphs at once instead of word by word](https://groundtruth.day/news/text-that-arrives-all-at-once.html) (2026-06-22) — Diffusion text models generate in parallel blocks rather than left to right; Google's open DiffusionGemma and Inception's Mercury 2 are now in a head-to-head over speed. · source: https://huggingface.co/google/diffusiongemma-26B-A4B-it · data: https://groundtruth.day/news/text-that-arrives-all-at-once.json
- [A big study finds AI more persuasive than professional human persuaders](https://groundtruth.day/news/ai-can-out-talk-the-professionals.html) (2026-06-22) — Across roughly nineteen thousand real conversations, AI systems drove far more charitable donations than trained human canvassers -- shifting the question to 'on whose behalf.' · source: https://jack-clark.net · data: https://groundtruth.day/news/ai-can-out-talk-the-professionals.json
- [A trust wobble hits AI coding tools: hidden reasoning and a runaway bug](https://groundtruth.day/news/can-you-trust-what-the-coding-agent-tells-you.html) (2026-06-22) — Two heated developer threads converge on one worry -- whether you can trust what an AI coding assistant shows you it's thinking, and what it quietly does to your machine. · source: https://github.com/openai/codex/issues/28224 · data: https://groundtruth.day/news/can-you-trust-what-the-coding-agent-tells-you.json
- [A tiny image-editing AI now runs entirely inside your web browser](https://groundtruth.day/news/a-tiny-image-fixer-that-runs-in-your-browser.html) (2026-06-22) — Moebius is a small inpainting model claiming far-larger-model quality, and a developer ported it to run on your own machine in a browser tab -- no server, no upload. · source: https://simonwillison.net/2026/Jun/22/porting-moebius/ · data: https://groundtruth.day/news/a-tiny-image-fixer-that-runs-in-your-browser.json
- [Google DeepMind puts $75 million into film studio A24 to build AI moviemaking tools](https://groundtruth.day/news/google-deepmind-bets-on-a-film-studio.html) (2026-06-22) — A frontier AI lab is investing in a prestige studio to develop production tools hands-on with filmmakers -- officially not a deal to train models on A24's films. · source: https://deadline.com/2026/06/google-a24-partnership-ai-filmmaking-tools/ · data: https://groundtruth.day/news/google-deepmind-bets-on-a-film-studio.json
- [The best free AI model just landed — but almost nobody can run it at home](https://groundtruth.day/news/open-license-closed-hardware.html) (2026-06-21) — A powerful open model anyone can legally download has reignited the open-vs-closed debate — but it's so large that 'open' now means 'open if you own a small server.' · source: https://huggingface.co/zai-org/GLM-5.2 · data: https://groundtruth.day/news/open-license-closed-hardware.json
- [A 61-author paper argues AI leaderboards quietly mislead everyone](https://groundtruth.day/news/the-leaderboard-is-lying.html) (2026-06-21) — A large industry-led study makes a blunt case: the rankings everyone cites to pick the 'best' AI agent don't survive contact with the real world. · source: https://arxiv.org/abs/2606.19704 · data: https://groundtruth.day/news/the-leaderboard-is-lying.json
- [A robot hand learns to open things by reasoning about touch, not video](https://groundtruth.day/news/robot-hands-that-feel-the-handle.html) (2026-06-21) — New research teaches multi-finger robot hands to manipulate things with moving parts — handles, drawers, hinges — by focusing on contact points, and stays steady even without touch sensors. · source: https://arxiv.org/abs/2606.15133 · data: https://groundtruth.day/news/robot-hands-that-feel-the-handle.json
- [An image generator that catches and corrects its own errors mid-draw](https://groundtruth.day/news/models-that-fix-their-own-mistakes.html) (2026-06-21) — Image-generating models often quietly break the very rule they were told to follow. A new method trains them to notice that error as they work and steer back on target. · source: https://arxiv.org/abs/2606.20404 · data: https://groundtruth.day/news/models-that-fix-their-own-mistakes.json
- [Researchers turn the internet's hobbyist art 'filters' into training fuel](https://groundtruth.day/news/community-styles-become-training-data.html) (2026-06-21) — Cleanly separating 'what's in a picture' from 'what style it's in' usually needs scarce data. A new method mines the huge public library of community-made style add-ons instead. · source: https://arxiv.org/abs/2606.20506 · data: https://groundtruth.day/news/community-styles-become-training-data.json
- [AI builds a single 3D object that shows two different things from two angles](https://groundtruth.day/news/one-object-two-pictures.html) (2026-06-21) — A new training-free method generates 3D visual illusions — one sculpture that reads as completely different objects depending on where you stand — in minutes instead of hours. · source: https://arxiv.org/abs/2606.20563 · data: https://groundtruth.day/news/one-object-two-pictures.json
- [When an AI assistant hides a glitch by inventing a story](https://groundtruth.day/news/the-error-that-becomes-a-story.html) (2026-06-20) — Researchers watched a real AI assistant for two months and found its scariest failures weren't crashes — they were confident, made-up explanations built on top of errors it quietly swallowed. · source: https://arxiv.org/abs/2606.14589 · data: https://groundtruth.day/news/the-error-that-becomes-a-story.json
- [AI 'world models' have short-term memory — they forget what's off-screen](https://groundtruth.day/news/the-room-resets-when-you-look-away.html) (2026-06-20) — A sweeping study of dozens of AI video-prediction systems finds they don't truly remember the world; when something leaves the frame, they quietly reinvent it the next time you look. · source: https://huggingface.co/papers/2606.20545 · data: https://groundtruth.day/news/the-room-resets-when-you-look-away.json
- [A world model that thinks in loops instead of stacking layers](https://groundtruth.day/news/one-block-thinking-in-loops.html) (2026-06-20) — Instead of building an ever-deeper neural network to simulate the future, a new design re-runs one small block over and over — doing comparable work with a fraction of the size. · source: https://arxiv.org/abs/2606.18208 · data: https://groundtruth.day/news/one-block-thinking-in-loops.json
- [Robots may not need to picture the future as video to act on it](https://groundtruth.day/news/robots-that-dont-need-to-imagine-video.html) (2026-06-20) — Generating a full imagined video of what comes next is expensive. A new method skips it — pulling a robot's next move straight from the inner workings of an image-editing model. · source: https://huggingface.co/papers/2606.19531 · data: https://groundtruth.day/news/robots-that-dont-need-to-imagine-video.json
- [Teaching AI with rewards — minus the expensive second model that grades it](https://groundtruth.day/news/reward-training-without-a-referee.html) (2026-06-20) — The standard way to polish a model with rewards quietly runs a second 'critic' model alongside it. A new method derives the critic's judgment from the model itself, dropping the extra cost. · source: https://arxiv.org/abs/2606.20008 · data: https://groundtruth.day/news/reward-training-without-a-referee.json
- [An openly-released text model that writes by refining, not word-by-word](https://groundtruth.day/news/a-bigger-text-model-that-doesnt-write-left-to-right.html) (2026-06-20) — Most language models write one word after another, left to right. A new openly-released model of real size generates text the way image AIs make pictures — refining a whole draft at once. · source: https://huggingface.co/papers/2606.19005 · data: https://groundtruth.day/news/a-bigger-text-model-that-doesnt-write-left-to-right.json
- [An AI agent design that refuses to act on what it merely assumes](https://groundtruth.day/news/an-agent-that-only-trusts-what-it-sees.html) (2026-06-20) — Tool-using agents often act on what they think is true rather than what they've checked. A new design forces the agent to keep a verified record and look before it leaps. · source: https://huggingface.co/papers/2606.20529 · data: https://groundtruth.day/news/an-agent-that-only-trusts-what-it-sees.json
- [AI coding skill in Python doesn't carry over to other languages](https://groundtruth.day/news/good-at-python-isnt-good-at-coding.html) (2026-06-20) — A widely-trusted coding benchmark was Python-only. Expanding it to a dozen languages revealed that models acing Python often stumble badly elsewhere — Python skill isn't general coding skill. · source: https://huggingface.co/papers/2606.20517 · data: https://groundtruth.day/news/good-at-python-isnt-good-at-coding.json
- [Independent testers probed the labs' secret models — and graded the danger](https://groundtruth.day/news/safety-testers-get-inside-the-frontier-labs.html) (2026-06-20) — A safety group got rare access to unreleased AI agents inside the top labs. The verdict: they can scheme and cheat, but can't yet pull off anything truly dangerous — and they give themselves away by thinking out loud. · source: https://metr.org/blog/2026-05-19-frontier-risk-report/ · data: https://groundtruth.day/news/safety-testers-get-inside-the-frontier-labs.json
- [Polishing AI by looking inside its 'mind' instead of just thumbs-up, thumbs-down](https://groundtruth.day/news/shaping-the-reward-by-looking-inside.html) (2026-06-20) — Reward training usually treats the model as a black box — thumbs up, thumbs down, hope for the best. A new method peers inside to see why an answer was preferred, and shapes the lesson on purpose. · source: https://arxiv.org/abs/2606.12360 · data: https://groundtruth.day/news/shaping-the-reward-by-looking-inside.json
- [A powerful open model lands and reignites the open-vs-closed debate](https://groundtruth.day/news/glm-5-2-open-model-takes-on-the-giants.html) (2026-06-20) — A Chinese lab released a flagship model anyone can download and run, with a huge memory for long documents — and a viral claim that it makes things up less than a top closed model. · source: https://huggingface.co/zai-org/GLM-5.2-FP8 · data: https://groundtruth.day/news/glm-5-2-open-model-takes-on-the-giants.json
- [The hidden escape hatch in AI safety controls](https://groundtruth.day/news/safety-control-hidden-escape-hatch.html) (2026-06-19) — Researchers at Hong Kong Polytechnic University show that clamping an AI safety feature — like one that controls refusals — doesn't remove the behavior. It hides in the part of the model's internal state that the safety tool throws away, and can be recovered while the monitored feature looks perfectly controlled. · source: https://arxiv.org/abs/2606.18322 · data: https://groundtruth.day/news/safety-control-hidden-escape-hatch.json
- [Your AI judge might be reliable — and still be wrong](https://groundtruth.day/news/ai-judges-reliable-but-wrong.html) (2026-06-19) — The largest audit of AI language model judges to date — 21 judges, over half a million grading decisions — finds that standard reliability metrics are inflated by roughly a third, that the same judge can score differently on different benchmarks, and that high consistency and severe bias can coexist in the same system. · source: https://arxiv.org/abs/2606.19544 · data: https://groundtruth.day/news/ai-judges-reliable-but-wrong.json
- [Turn the camera away, and the AI's world freezes](https://groundtruth.day/news/world-models-camera-turns-world-freezes.html) (2026-06-19) — A new benchmark tests whether video AI systems can track what happens to parts of a scene the camera isn't currently showing. Across 23 models, the answer is mostly no — and making the models larger made the problem worse, not better. · source: https://arxiv.org/abs/2606.20545 · data: https://groundtruth.day/news/world-models-camera-turns-world-freezes.json
- [A robot that runs its own experiments — and sometimes fails when it matters](https://groundtruth.day/news/robots-run-experiments-themselves.html) (2026-06-19) — NVIDIA researchers gave AI coding agents full control of a physical robot lab — including automated reset and vision-based success checking. One agent inserted a graphics card into a motherboard. The headline success rate is real but requires a close read. · source: https://research.nvidia.com/labs/gear/enpire/ · data: https://groundtruth.day/news/robots-run-experiments-themselves.json
- [A tiny image-fixer keeps up with a model fifty times its size](https://groundtruth.day/news/tiny-image-fixer-beats-a-giant.html) (2026-06-19) — Filling in the missing parts of an image usually takes a huge model. This one is a small fraction of the size and far faster, yet matches a system far bigger than it. · source: https://arxiv.org/abs/2606.19195 · data: https://groundtruth.day/news/tiny-image-fixer-beats-a-giant.json
- [What if a word were a rotation? A more mathematical way to build AI](https://groundtruth.day/news/words-as-rotations.html) (2026-06-19) — A fresh, abstract idea: treat what a model attends to not as plain lists of numbers but as geometric moves like rotations — so useful symmetries come 'for free.' Elegant and early. (A deeper, technical read.) · source: https://arxiv.org/abs/2606.20547 · data: https://groundtruth.day/news/words-as-rotations.json
- [Faster AI training by quietly cloning the model](https://groundtruth.day/news/faster-training-by-cloning-the-model.html) (2026-06-19) — Teaching a model with rewards is slow because it has to write out endless practice answers. A new trick: make a cheap, shrunk-down copy of the model to crank those out faster. · source: https://arxiv.org/abs/2606.18967 · data: https://groundtruth.day/news/faster-training-by-cloning-the-model.json
- [An AI that could rewrite its own words — and gained nothing from it](https://groundtruth.day/news/the-ai-that-could-edit-itself-but-didnt.html) (2026-06-19) — A different style of text AI can go back and change any word at any point as it writes. Given that power, it didn't actually produce better writing. A clean negative result. · source: https://arxiv.org/abs/2606.19005 · data: https://groundtruth.day/news/the-ai-that-could-edit-itself-but-didnt.json
- [Crediting an AI for the right steps — without a second model to judge them](https://groundtruth.day/news/credit-without-a-critic.html) (2026-06-19) — When you reward an AI for a good final answer, it's hard to know which of its steps earned the credit. The usual fix is training a second 'judge' model. This skips that. · source: https://arxiv.org/abs/2606.20008 · data: https://groundtruth.day/news/credit-without-a-critic.json
- [Giving an AI real spatial tools instead of letting it guess](https://groundtruth.day/news/ai-that-uses-spatial-tools-instead-of-guessing.html) (2026-06-19) — Vision AIs are surprisingly bad at precise 'where is this in 3D space' questions. This one stops guessing and calls dedicated spatial tools, while keeping a memory across views. · source: https://arxiv.org/abs/2606.20515 · data: https://groundtruth.day/news/ai-that-uses-spatial-tools-instead-of-guessing.json
- [Do robots even need to imagine the movie?](https://groundtruth.day/news/robots-imagine-one-frame.html) (2026-06-19) — The common belief is that a robot needs to imagine a video of what happens next to plan. A new method says no — imagine a single still frame, and don't even fully draw it. · source: https://arxiv.org/abs/2606.19531 · data: https://groundtruth.day/news/robots-imagine-one-frame.json
- [Reliable, and still wrong](https://groundtruth.day/news/reliable-but-wrong-judges.html) (2026-06-19) — Using one AI to grade another is now common — but the biggest audit yet shows these graders are consistent without being correct. A judge that always picks "answer A" scores perfectly on consistency. · source: https://arxiv.org/abs/2606.19544 · data: https://groundtruth.day/news/reliable-but-wrong-judges.json
- [A coding assistant ran a real robot](https://groundtruth.day/news/coding-agent-robot.html) (2026-06-19) — An AI coding agent read the research, wrote the control code, watched it fail, and fixed it — seating a graphics card into a motherboard by itself. The honest catch: most of the success is retrying. · source: https://research.nvidia.com/labs/gear/enpire/ · data: https://groundtruth.day/news/coding-agent-robot.json
- [The little words that keep AI from getting boring](https://groundtruth.day/news/forking-words.html) (2026-06-19) — Rewarding a reasoning model too hard makes it repetitive — and the casualties are tiny words like "but" and "instead" that let it branch to a better thought. A near-free fix protects them. · source: https://arxiv.org/abs/2606.19236 · data: https://groundtruth.day/news/forking-words.json
- [Turn around, and the world disappears](https://groundtruth.day/news/world-models-forget.html) (2026-06-19) — AI video models that are supposed to "understand" a 3D scene only remember what's on screen — pan away and back, and things have reset. Bigger models are worse at it. · source: https://arxiv.org/abs/2606.20545 · data: https://groundtruth.day/news/world-models-forget.json
- [The safety switch that doesn't actually work](https://groundtruth.day/news/sae-safety-switch.html) (2026-06-19) — A control that's supposed to force an AI to refuse harmful requests gets bypassed while it's switched on — the bad behavior hides in the part of the tool that gets thrown away. · source: https://arxiv.org/abs/2606.18322 · data: https://groundtruth.day/news/sae-safety-switch.json

## Learn (curated, cited lessons)
- [Prompt injection: the con that hijacks AI agents](https://groundtruth.day/learn/prompt-injection.html) — Prompt injection is when hidden instructions in the content an AI reads trick it into ignoring its real orders, the core security problem of any AI that browses, reads email, or uses a computer. · data: https://groundtruth.day/learn/prompt-injection.json
- [Distillation: how a small AI learns from a big one](https://groundtruth.day/learn/distillation.html) — Distillation trains a smaller, cheaper model to imitate a larger, smarter one, the idea behind both efficient deployment and the 'copying' accusations now driving AI geopolitics. · data: https://groundtruth.day/learn/distillation.json
- [Synthetic Data: When AI Makes Its Own Training Material](https://groundtruth.day/learn/synthetic-data.html) — The internet is running out of fresh text to train on, so the most advanced models increasingly learn from data that other AI made or shaped. Here is how that works, why it helps, and how it can quietly poison a model. · data: https://groundtruth.day/learn/synthetic-data.json
- [Mixture of Experts: The Committee Inside a Giant Model](https://groundtruth.day/learn/mixture-of-experts.html) — Why the biggest AI models are not really one big brain but a large team of specialists, only a few of whom wake up for any given word -- the trick that lets a model be huge and fast at the same time. · data: https://groundtruth.day/learn/mixture-of-experts.json
- [Recursive self-improvement: when AI starts building AI](https://groundtruth.day/learn/recursive-self-improvement.html) — The idea that an AI good enough at AI research could improve itself, and the improved version could improve itself again, faster each round. Here's what it actually means, why a major lab now says we're getting close, and why "close" is not the same as "here." · data: https://groundtruth.day/learn/recursive-self-improvement.json
- [AI Persuasion: When Machines Get Good at Changing Your Mind](https://groundtruth.day/learn/ai-persuasion.html) — Why language models have quietly become powerful persuaders, how they do it, and why researchers treat 'superpersuasion' as a safety problem rather than a marketing feature. · data: https://groundtruth.day/learn/ai-persuasion.json
- [How AI Gets Benchmarked — and Why the Leaderboard Can Lie](https://groundtruth.day/learn/how-ai-is-benchmarked.html) — Every 'this AI is now #1' headline rests on a benchmark. Here's how those tests actually work, why a top score doesn't always mean what you think, and how to read a leaderboard like a skeptic. · data: https://groundtruth.day/learn/how-ai-is-benchmarked.json
- [Open vs. closed AI models — what "open weights" really means](https://groundtruth.day/learn/open-weight-models.html) — Some AI models you can only rent through a company's interface; others you can download and run yourself. That difference — open weights vs. closed — shapes privacy, research, cost, and who controls the technology. · data: https://groundtruth.day/learn/open-weight-models.json
- [Scaling laws — does bigger always mean better?](https://groundtruth.day/learn/scaling-laws.html) — For years, AI progress ran on a simple recipe: make the model bigger, feed it more data, get a better model. That pattern is real and predictable — but it has limits and surprises. Here's what scaling laws actually say. · data: https://groundtruth.day/learn/scaling-laws.json
- [What is a context window?](https://groundtruth.day/learn/context-windows.html) — A model's context window is how much text it can hold in mind at once — its working memory. Bigger is useful, but a long window isn't the same as a good memory. Here's how it works and where it breaks. · data: https://groundtruth.day/learn/context-windows.json
- [Why does AI make things up?](https://groundtruth.day/learn/hallucination.html) — Language models sometimes state false things with total confidence — a behavior called hallucination. It isn't a bug they'll simply patch out; it falls out of how they're built. Here's why it happens and how people fight it. · data: https://groundtruth.day/learn/hallucination.json
- [What makes an AI an "agent"?](https://groundtruth.day/learn/ai-agents.html) — An AI agent doesn't just answer questions — it takes actions: calling tools, running steps, and reacting to what it finds. Here's the loop at the core of every agent, and why agents fail in their own peculiar ways. · data: https://groundtruth.day/learn/ai-agents.json
- [What does it mean for AI to grade AI?](https://groundtruth.day/learn/llm-as-a-judge.html) — We increasingly use one AI model to evaluate another's answers — because human grading doesn't scale. Here's how 'AI as a judge' works, why it's everywhere, and the traps that make it unreliable. · data: https://groundtruth.day/learn/llm-as-a-judge.json
- [Mechanistic interpretability & sparse autoencoders](https://groundtruth.day/learn/mechanistic-interpretability.html) — What people mean by "reading a model's mind" — finding human-understandable features inside a neural network, the tools that do it, and where those tools fall short. · data: https://groundtruth.day/learn/mechanistic-interpretability.json
- [Reward-based fine-tuning (RLHF and RLVR)](https://groundtruth.day/learn/rl-post-training.html) — After a model is first trained, it gets "polished" by rewarding good answers. Here's what that phase is, why it works, and the failure mode where models get repetitive and dull. · data: https://groundtruth.day/learn/rl-post-training.json
- [What are diffusion language models?](https://groundtruth.day/learn/diffusion-language-models.html) — Most AI writes one word at a time and can never go back. Diffusion language models start from noise and clarify it iteratively — and some versions can revise any word at any step. A growing alternative to the standard left-to-right approach. · data: https://groundtruth.day/learn/diffusion-language-models.json
- [What are world models?](https://groundtruth.day/learn/world-models.html) — A world model is an AI system's internal understanding of how an environment works — not just what it sees right now, but what will happen after an action, and what would have happened differently. Central to planning, robotics, and the next generation of physical AI. · data: https://groundtruth.day/learn/world-models.json

## Tools (usable, shipping AI tools)
- [Kimi (Kimi K2.6)](https://www.kimi.com) — Moonshot AI's web assistant and agent, running the open-weight Kimi K2.6 model; free to use in the browser for chat and long-horizon agent tasks, with the weights also downloadable for self-hosting. [AI assistant / coding agent]
- [Modular MAX + Mojo](https://www.modular.com/) — A programming language (Mojo) and compiler/runtime (MAX) for running AI models efficiently across different hardware instead of being locked to one chip vendor; now being acquired by Qualcomm but still openly available to developers. [AI compiler / runtime]
- [Gemma-4 WebGPU Kernels](https://huggingface.co/spaces/webml-community/Gemma-4-WebGPU-Kernels) — A demo running Google's Gemma-4 model directly inside a web browser using your device's graphics hardware — private, on-device AI with no server and no data leaving your machine. [AI in the browser]
- [OpenMontage](https://github.com/calesthio/OpenMontage) — An open-source system that turns an AI coding assistant into an automated video-production studio, with a large library of pipelines, tools, and agent skills for editing and assembling video. [AI video production]
- [Gemini 3.5 Flash computer use](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-computer-use-gemini-3-5-flash/) — Google's fast model can now operate a browser, phone, or desktop directly as a built-in tool, with optional confirm-before-acting and auto-stop-on-attack safeguards for building automation agents. [Agent / automation]
- [Cloudflare Temporary Accounts](https://blog.cloudflare.com/temporary-accounts) — Lets an automated agent deploy and run on Cloudflare before a human signs up, removing the account-creation step from agent workflows. [Agent deployment infra]
- [DeerFlow](https://github.com/bytedance/deer-flow) — ByteDance's open-source agent harness that breaks a long task into specialist sub-agents running in parallel, executes code safely in sandboxes, keeps memory across sessions, and produces reports, slides, and pages; built on LangChain and works with multiple model providers. [Agent framework]
- [NVIDIA SkillSpector](https://github.com/NVIDIA/skillspector) — A scanner that inspects agent skills for security problems before you run them -- a static safety check for the fast-growing agent-skill supply chain. [Agent security scanner]
- [RAGFlow](https://github.com/infiniflow/ragflow) — An open engine for building AI question-answering over your own files and documents. [Build with your own documents]
- [Claude Code](https://www.anthropic.com/claude-code) — Anthropic's command-line coding agent that reads a whole codebase, edits files, runs tests and fixes failures on its own; it is the tool behind Anthropic's disclosure that Claude now authors most of its production code. [Coding agent]
- [ComfyUI](https://github.com/comfyanonymous/ComfyUI) — A visual, node-based studio for generating images and video with open models. Powerful and endlessly extensible. [Create images & video]
- [Headroom](https://github.com/chopratejas/headroom) — A drop-in proxy that sits between your coding assistant and the AI model and automatically compresses bulky tool outputs, logs, and retrieved text before they reach the model — cutting token usage sharply without changing your code. [Cut AI agent costs]
- [Mercury 2 (Inception Labs)](https://inceptionlabs.ai) — An API-only diffusion language model pitched on raw speed, claiming to out-pace open diffusion models on tokens-per-second for latency-sensitive generation. [Diffusion LLM API]
- [Claude Tag (agent identity access model)](https://claude.com/blog/agent-identity-access-model) — Anthropic's product for putting Claude to work in shared team channels, now with an access model that gives each agent its own scoped accounts in the systems it touches -- GitHub, Slack, a data warehouse -- instead of borrowing an individual user's permissions, so every action is bounded and audited. [Enterprise agent platform]
- [Hugging Face](https://huggingface.co) — The main hub for finding, downloading, and trying open AI models and datasets — the field's town square. [Find models & datasets]
- [TimesFM](https://github.com/google-research/timesfm) — Google's pre-trained foundation model for time-series forecasting — predicting things that change over time, like demand, traffic, or sensor readings — usable out of the box without training your own model. [Forecasting model]
- [codebase-memory-mcp](https://github.com/DeusData/codebase-memory-mcp) — Indexes an entire codebase into a persistent, queryable knowledge graph so AI agents can understand large projects fast. Supports a huge range of programming languages, answers queries near-instantly, and ships as a single dependency-free binary. [Give AI agents code memory]
- [GLM-5.2 on Baseten](https://www.baseten.co/blog/how-we-built-the-worlds-fastest-api-for-glm-52/) — The top trending open-weight model served as a fast hosted endpoint, reported at 280+ tokens/sec on Blackwell-class hardware -- an open model you can call like a closed one. [Hosted open-model API]
- [Gemma-4 12B Coder (GGUF)](https://huggingface.co/yuxinlu1/gemma-4-12B-coder-fable5-composer2.5-v1-GGUF) — A fine-tuned, locally-runnable version of Google's Gemma-4 model specialized for programming tasks, packaged in a format that runs efficiently on everyday consumer hardware. [Local coding model]
- [Skybridge](https://www.producthunt.com/posts/skybridge) — A framework for building MCP-native apps -- interactive tools an AI assistant can open and use directly, pitched as 'MCP apps are the new website.' [MCP app framework]
- [Sakana Fugu](https://sakana.ai/fugu/) — A single OpenAI-compatible endpoint that dynamically routes each request across several frontier models, so you call one API and get a coordinated multi-model answer. [Model-orchestration API]
- [LLaDA / iLLaDA](https://github.com/ML-GSAI/LLaDA) — An openly released diffusion language model (weights and code) that generates text by refining a whole passage at once rather than one word at a time, useful for experimenting with non-autoregressive generation and infilling. [Open language model]
- [GLM-5.2](https://huggingface.co/zai-org/GLM-5.2-FP8) — A flagship openly-available language model with a very large context window for long documents and code. Free to download and run yourself, with compressed versions for more modest hardware. [Open large language model]
- [Kimi K2.6 weights (Hugging Face)](https://huggingface.co/moonshotai/Kimi-K2.6) — The actual Kimi K2.6 model weights, published under a modified-MIT license for anyone to download, run, and build on; large enough that full-strength use needs a multi-GPU node. [Open model download]
- [MiniMax-M3](https://huggingface.co/MiniMaxAI/MiniMax-M3) — A natively multimodal open model trained on text, image, and video from the first step, with a million-token context and a sparse-attention design built for speed; downloadable for self-hosting and also offered through MiniMax's own API and agent platform. [Open-weight model]
- [Qwen-AgentWorld](https://github.com/QwenLM/Qwen-AgentWorld) — Alibaba's open language world model that simulates agent environments -- browser, terminal, phone, coding workspace and more -- so other agents can be trained inside the simulation. Released with open weights and code in two sizes. [Open-weight model (agent world model)]
- [DiffusionGemma](https://huggingface.co/google/diffusiongemma-26B-A4B-it) — Google's open-weight text-diffusion model that generates text in parallel blocks instead of one token at a time; Apache-2.0, runnable locally, with community tooling already shipping. [Open-weight model (self-host)]
- [SGLang v0.5.13](https://github.com/sgl-project/sglang/releases) — A high-performance open serving engine for language models. The new version turns on faster 'guess-ahead' decoding by default and trims scheduling overhead for quicker responses. [Run AI models efficiently]
- [vLLM v0.23.0](https://github.com/vllm-project/vllm/releases) — The widely-used open engine for serving language models fast and cheaply. The latest release adds smarter memory handling for long conversations and faster GPU execution. [Run AI models efficiently]
- [LM Studio](https://lmstudio.ai) — A friendly desktop app to find, download, and chat with open models on your own machine — no command line needed. [Run models on your computer]
- [Ollama](https://ollama.com) — Download and run open AI models locally with a single command. The easiest on-ramp to running your own model. [Run models on your computer]
- [Open WebUI](https://github.com/open-webui/open-webui) — A polished, ChatGPT-style web interface for the open models you run yourself. [Run models on your computer]
- [llama.cpp](https://github.com/ggml-org/llama.cpp) — The lean, fast engine that makes big models run on ordinary laptops; powers much of the local-AI ecosystem. [Run models on your computer]
- [OpenAI Codex Security (Daybreak)](https://openai.com/index/patch-the-planet/) — An in-IDE plugin from OpenAI's Daybreak initiative that finds, validates, and fixes software vulnerabilities, plus an open-source remediation program run with Trail of Bits and HackerOne. [Security coding assistant]
- [vLLM](https://github.com/vllm-project/vllm) — The popular open engine for serving AI models fast and efficiently when you need to handle real traffic. [Serve at scale]
- [veRL](https://github.com/volcengine/verl) — The open RL post-training framework used by most research labs training reasoning models today. Run GRPO, PPO, and related reward-training methods on your own models. [Train & fine-tune AI models]

## Hackathons (remote/online AI hackathons)
- [Hack Begin](https://hack-begin.devpost.com/) — deadline May 19 - Jun 25, 2026; 1 non-cash prize (swag/credits/recognition) · Online — fully remote via Devpost
- [Google AI Workshop Hackathon](https://google-ai-workshop-hackathon.devpost.com/) — deadline Jun 24 - 26, 2026; 3 non-cash prizes (swag/credits/recognition) · Online — fully remote via Devpost
- [Built with Python Hackathon](https://built-with-python-hackathon.devpost.com/) — deadline Jun 06 - 27, 2026; 5 non-cash prizes (swag/credits/recognition) · Online — fully remote via Devpost
- [Hack Munch ](https://hack-munch-30312.devpost.com/) — deadline Jun 03 - 28, 2026; 1 non-cash prize (swag/credits/recognition) · Online — fully remote via Devpost
- [Hack Verse](https://hack-verse-30300.devpost.com/) — deadline Jun 02 - 28, 2026; 1 non-cash prize (swag/credits/recognition) · Online — fully remote via Devpost
- [Hack Verse](https://hack-verse-30325.devpost.com/) — deadline Jun 04 - 28, 2026; 1 non-cash prize (swag/credits/recognition) · Online — fully remote via Devpost
- [Hack-Vserse](https://hack-vserse.devpost.com/) — deadline Jun 03 - 28, 2026; 1 non-cash prize (swag/credits/recognition) · Online — fully remote via Devpost
- [HackNova](https://hacknova-30322.devpost.com/) — deadline Jun 04 - 28, 2026; 1 non-cash prize (swag/credits/recognition) · Online — fully remote via Devpost
- [UiPath AgentHack](https://uipath-agenthack.devpost.com/) — deadline May 15 - Jun 29, 2026; $50,000 across 16 cash prizes · Online — fully remote via Devpost
- [305 HackShells Edition June 2026 ](https://305hackshellsjune2026.devpost.com/) — deadline Jun 01 - 30, 2026; 1 non-cash prize (swag/credits/recognition) · Online — fully remote via Devpost
- [Build the Future with AI — From Code to No-Code](https://build-the-future-with-ai.devpost.com/) — deadline May 31 - Jun 30, 2026; 7 non-cash prizes (swag/credits/recognition) · Online — fully remote via Devpost
- [Moonshot Hackathon](https://moonshot-aethra.devpost.com/) — deadline Jun 03 - 30, 2026; $33,532 across 7 cash prizes · Online — fully remote via Devpost
- [Impact Creation](https://impact-creation.devpost.com/) — deadline Jun 13 - Jul 01, 2026; 1 non-cash prize (swag/credits/recognition) · Online — fully remote via Devpost
- [SunnyHacks June 2026](https://sunnyhacks-june-2026.devpost.com/) — deadline Jun 01 - Jul 01, 2026; 2 non-cash prizes (swag/credits/recognition) · Online — fully remote via Devpost
- [FutureAI Global Hackathon 2026](https://futureai-global-hackthon.devpost.com/) — deadline May 29 - Jul 05, 2026; 6 non-cash prizes (swag/credits/recognition) · Online — fully remote via Devpost
- [Global AI Hackathon Series with Qwen Cloud ](https://qwencloud-hackathon.devpost.com/) — deadline May 26 - Jul 09, 2026; $45,000 across 7 cash prizes · Online — fully remote via Devpost
- [Kaya AI IIT India Hackathon 2026](https://kaya-ai-iit-hackathon-2026.devpost.com/) — deadline Jun 10 - Jul 10, 2026; ₹ 350,000 across 3 cash prizes · Online — fully remote via Devpost
- [LUMA Hackathon (July 3rd - 10th)](https://luma-hackathon-500.devpost.com/) — deadline Apr 11 - Jul 10, 2026; 1 non-cash prize (swag/credits/recognition) · Online — fully remote via Devpost
- [CROO Agent Hackathon](https://dorahacks.io/hackathon/croo-hackathon) — deadline Jul 12, 2026; $10,200 USD split across 6 AI agent tracks · Fully online on DoraHacks; open to all participants worldwide
- [MLH Global Hack Week: Season Launch](https://ghw.mlh.com/events/season-launch) — deadline Jul 16, 2026; Live challenges with swag and recognition; no cash prizes · 100% online and free for anyone anywhere; community via Discord
- [Hoobit Hacks 2026](https://hoobit-hacks-2026.devpost.com/) — deadline Mar 30 - Jul 18, 2026; 2 non-cash prizes (swag/credits/recognition) · Online — fully remote via Devpost
- [BuunieX Hackathon](https://buuniex-hackathon.devpost.com/) — deadline Jun 22 - Jul 22, 2026; 1 non-cash prize (swag/credits/recognition) · Online — fully remote via Devpost
- [India High School Exoplanet Data Challenge](https://celesta-exoplanet-challenge.devpost.com/) — deadline Jun 15 - Jul 31, 2026; $10,300 across 2 cash prizes · Online — fully remote via Devpost
- [Backblaze Generative Media Hackathon: Build with Genblaze on B2](https://backblaze-generative-media.devpost.com/) — deadline Jun 22 - Aug 03, 2026; $10,000 across 3 cash prizes · Online — fully remote via Devpost
- [IncludAI: The Neurodiversity Hackathon ](https://includai--2026.devpost.com/) — deadline Jun 24 - Aug 09, 2026; $3,000 across 5 cash prizes · Online — fully remote via Devpost
- [MLH Global Hack Week: Agents](https://ghw.mlh.com/events/agents) — deadline Aug 13, 2026; Live challenges with swag and recognition; no cash prizes · 100% online and free for anyone anywhere; community via Discord
- [Arm Create: AI Optimization Challenge](https://arm-ai-optimization-challenge.devpost.com/) — deadline Jun 04 - Aug 14, 2026; $8,000 across 5 cash prizes · Online — fully remote via Devpost
- [AceSAT Education AI-Agent](https://acesat-ai-agent.devpost.com/) — deadline Jun 12 - Aug 15, 2026; $100 across 1 cash prize · Online — fully remote via Devpost
- [Build with Gemini XPRIZE](https://xprize.devpost.com/) — deadline May 19 - Aug 17, 2026; $2,000,000 across 11 cash prizes · Online — fully remote via Devpost
- [NeuralSprint](https://neuralsprint.devpost.com/) — deadline Jun 18 - Aug 24, 2026; 5 non-cash prizes (swag/credits/recognition) · Online — fully remote via Devpost
- [Africa Deep Tech Challenge 2026](https://adtc-2026.devpost.com/) — deadline Jun 17 - Aug 25, 2026; $16,500 across 4 cash prizes · Online — fully remote via Devpost
- [AI x City Climate Action Hackathon 2026](https://www.innovate4cities.org/hackathon/hackathon2026/) — deadline Aug 31, 2026; City implementation partnership; top 10 pitch at Cambridge; travel support for top 3 to UTM Global Innovation Summit in Spain (Nov 3) · Online submissions open globally; optional in-person pitching finale at University of Cambridge in September
- [AI YES :International Youth AI Competition](https://ai-yes-competition-30441.devpost.com/) — deadline Jun 17 - Sep 01, 2026; 3 non-cash prizes (swag/credits/recognition) · Online — fully remote via Devpost
- [VoltHacks](https://volthacks.devpost.com/) — deadline May 22 - Sep 05, 2026; $2,905 across 4 cash prizes · Online — fully remote via Devpost
- [AI GENESIS 2026](https://lablab.ai/ai-hackathons/ai-genesis-2026) — deadline Nov 2, 2026; TBA; global grand prize with on-stage pitching at function1 Conference in Dubai · Hybrid: online build and collaboration phase Oct 26-Nov 2 open globally; optional in-person finale in Dubai Nov 3
- [DEMOKHE](https://demokhe.devpost.com/) — deadline Mar 24, 2026 - Mar 24, 2030; 1 non-cash prize (swag/credits/recognition) · Online — fully remote via Devpost