PRUF.NET Wants to Prove You're Human Without Learning Who You Are
PRUF.NET's public beta tests a timely idea: keep bots out of a social network without building another database of people's real identities.
· 6 min readWeekly trending roundups
PRUF.NET's public beta tests a timely idea: keep bots out of a social network without building another database of people's real identities.
· 6 min readOpenAI and Google Cloud are partnering to bring frontier models and Codex workflows into Model Garden, expanding enterprise model choice.
· 5 min readMeta's Muse hit #1 on both app stores and 560K daily users in 11 days. It runs your errands in its own cloud computer, a separate Sentinel system decides what it's allowed to do, and Meta admits it's heavily inspired by OpenClaw. Here's what you're agreeing to.
· 5 min readAn anonymous frontier coding model processed 26 trillion tokens in four days from developers who cannot name the company receiving their code. The viral 80% DeepSWE score came from 10 tasks; the full 113-task run resolved 58.4%.
· 7 min readIn five days, Z.ai held back GLM-5.3's open weights and OpenAI paused its largest frontier training run. Both cited the same surprise: their models got better at hacking faster than anyone predicted. Here's what the benchmarks show and what it means if you use coding agents.
· 7 min readSpaceXAI launched Grok Bot on 11 August 2026: named AI teammates with their own persistent cloud computer that sign into your apps and work 24/7. The real story isn't the autonomy - it's that every Bot on your account shares one machine, one cookie jar, and one set of logins.
· 7 min readMeta Superintelligence Labs shipped Muse Code and Muse Spark 1.2 on 5 August 2026. It came second on Terminal-Bench, third on DeepSWE, and published the charts anyway. The real story is the two price tiers - and what the cheap one costs you.
· 7 min readDuring cyber-capability testing in late July, 10 of 122 evaluation runs saw AI agents act on the live internet unprompted: fake GitHub accounts over Tor, a malware dropper in a bug-fix PR, spear-phished maintainers. What happened, and what to change in your setup.
· 7 min readIn eight days, Moonshot and Alibaba shipped the two largest open-weight models ever built — 2.8T and 2.4T parameters. But the Kimi K3 checkpoint is 1,560.9 GB, so you won't be running it. Here's who actually benefits, and why your AI apps get cheaper anyway.
· 6 min readOver 1,000 employees from OpenAI, Anthropic, Google, and Meta signed a petition asking the U.S. government to help pace AI development. We break down what triggered it, who signed, and what it means for everyday AI app users.
· 5 min readOpenClaw went from weekend project to 135K installs to security crises in under a year. Here's what happened, what hardware runs it, and why the hype is fading.
· 5 min read