Moonshot removed cyber-related data from Kimi K3.1 Pro as vulnerability research performance got better as post-training scaled

Sticker of the year:

‼️ Node.js has a Remote Code Execution vulnerability. CVE-2026-111497 is a high-severity bug in its Brotli decoder. Decompressing attacker-controlled data can leak memory or lead to remote code execution. You don’t need to use Brotli directly. Your app could hit this through fetch() or libraries like body-parser. OpenHack supply chain tracks this, patch now!

‼️ Patches for the PHP Vulnerability are out ‼️ Go ahead and patch your servers ASAP. Full article: openhack.com/blog/php2shell-…

DeepSeek v4.1 Flash is now on available on all OpenHack plans! 🎉

Ox Alpha is now available for FREE on OpenHack! Rumored to be an unreleased GLM model, it has no guardrails and has surpassed Fable 5 Max in coding and cybersecurity benchmarks. Use it for free all OpenHack plans until August 27th. Happy Hacking!

GLM 5.3 is now available on OpenHack Pro, Max and Enterprise. Happy Hacking!

‼️ ATTENTION ALL DEFCON ATTENDEES ‼️ Free OpenHack Pro for a month with code "DEFCON34" Thank you for your attention to this matter.

We're offering a whole range of open source models through OpenHack's inference - including all major open source models and OpenAI models. Free credits on signup ⚡️

📣 Introducing the all new OpenHack TUI - an open source terminal interface for security tasks, codebase scanning and pentesting. Connect any model. Scan & auto verify vulnerabilities. Pentest any app. No Guardrails. Live on openhack.com

Kimi K3 is 3x the pricing of Kimi K2.7🤯

welcome back Fable 5

Open source models are the only way forward - Cannot be banned - All restrictions can be removed in post training - Cannot be slowed down / degraded by the big labs - Self hosted, fully in house inference

We're in the same church ;) Added to our website and Github README

OpenHack just crossed a 100 stars on GitHub! 🎉 Thank you for your all your support!

Claude Fable 5 literally flags a simple port scan and switches back to Opus. This is why we're building OpenHack and betting hard on open source models.

We're giving away free credits for everyone to try out OpenHack! Install using `pipx install openhack` and signup to get started.

Introducing OpenHack. An Open Source Agentic Security Scanner that hunts and verifies vulnerabilities using open source models exclusively. Upto 40x cheaper, it is on par with Claude Opus 4.6 on CVE-Bench. Check it out at openhack.com!

Microsoft surpassed Claude Mythos using their new harness, MDASH (multi-model agentic scanning harness)! MDASH uses GPT-5.4, Claude Opus 4.6, Sonnet 4.6 and absolutely smashed it on CyberGym. microsoft.com/en-us/security…

❗️ShinyHunters have also removed Canvas from their extortion page. It is very likely they settled.

It’s a speculation. ShinyHunters removed Canvas / Instructure from their extortion page when the list of schools was also removed. Historically, nothing has really stopped ShinyHunters from leaking anything; they leak everything on their server based out of Russia: http://91.215.85.103/pay_or_leak/

ShinyHunters removed list of schools from their website. It seems like Instructure has privately negotiated with ShinyHunters and is working on getting Canvas back online.

🚨 BREAKING: Instructure, the company behind Canvas - the LMS tool used by almost every university in the United States, has been breached by popular threat actor ShinyHunters. List of breached schools: http://91.215.85.103/pay_or_leak/instructure_affected_schools_list.txt

`plain-crypto-js`, the underlying malicious dependency, has now been unpublished by @npmjs. Malicious versions `axios@1.14.1` and `axios@0.30.4` have also been unpublished