Chaining recon into exploitation is easy. Keeping every asset and finding straight is the hard part. ARTEX is an AI autonomous penetration testing system built as a multi-agent LLM with a Go backend and Next.js frontend. > A planner and workers share an asset graph and an exploration graph > Human-in-the-loop approval and full traffic recording > Assets can be synced directly from ScopeSentry by project or task > Ships as one binary or a Docker Compose stack Its README reports the default work agent count per task is 3, configurable in system settings. The UI listens on port 8787. ⭐ 1.7k stars in 14 days · Go · AGPL-3.0

300 followers 🎉 Thank you all. Next stop: 400.

Hitting a rate limit mid-session is easy. Knowing how much free capacity is left is the hard part. OmniRoute is a free MIT AI gateway that puts 375 providers and 1,400+ models behind one endpoint. > Quota-aware auto-fallback across 20 routing strategies > RTK + Caveman compression saving 15-95% tokens > Works with Claude Code, Codex, Cursor, Cline and Copilot > Catalogs 471 free-tier entries across 34 recurring pool keys Its README reports ~1.62B free tokens per month across third-party free tiers, with live used and remaining shown on its dashboard. ⭐ 80k stars · TypeScript · MIT

‼️ Five separate critical flaws landed across five developer ecosystems in a single day: PHP RCE, a Node.js Brotli compression RCE, a supply chain attack on LiteLLM, an HTTPS root certificate key leak at eMudhra, and a 0-click RCE in GitLab. Days earlier, a Linux KVM escape carried CVE-2026-99999. That is not a leak, a whistleblower, or a journalist's reconstruction. Every item is signed, dated, and public: a CVE number, a vendor advisory, a package registry entry. The problem is not that the information is hidden. It is that there is too much of it, arriving at once, and the reflex to skim past it is exactly the reflex an attacker is counting on. The question everyone asks is whether any single one of these is dangerous. That is the wrong question, and everyone already has an answer to it. The question nobody asks is what a defender is supposed to do when five unrelated systems fail in the same news cycle: triage by severity, or triage by what is actually reachable from where they sit. The LiteLLM compromise is the one worth slowing down on. LiteLLM is a Python library that proxies calls to language models, so it sits between an application and whichever model provider the application talks to. When a package like that is compromised, the malicious code does not need to exploit anything: it is installed on purpose, by a developer, with the application's own credentials in reach. The supply chain attack is not a bug in LiteLLM. It is LiteLLM being the delivery vehicle. A developer running LiteLLM in production now has to answer a question no CVE score answers: which version did my lockfile resolve, and when. A pinned hash catches this. A floating version range does not, and neither does a scanner that only looks at the version number after the fact. The same day's GitLab flaw is the opposite shape. A 0-click RCE means the attacker does not need the victim to open a file, click a link, or approve anything: reaching the service is enough. GitLab is the system that holds the source code, the CI configuration, and often the deployment credentials. A 0-click flaw there is not a bug in a tool. It is a bug in the room where all the other tools' keys are kept. CVE-2026-99999, the KVM escape, is the one that reads as trivia until it isn't. KVM is the component that lets one physical machine run many isolated virtual machines. An escape means code running inside a guest can reach the host, which is the boundary that every cloud provider's isolation story rests on. It did not happen today. It happened days ago, and it is already sliding out of the feed. Here is what the fatigue actually costs. Not the flaws that get missed because nobody read the advisory. The flaws that get missed because everybody read five advisories on the same morning and none of them got acted on. A defender who triages by what is reachable from their own infrastructure, rather than by what is loudest, is not being lazy. They are the only ones still reading by Thursday. The five items are not equally urgent for any given team. Two of them probably do not touch that team's stack at all. The work is knowing which two, and that work does not get done by scrolling faster. Early report: these facts come from first disclosures and are not confirmed yet. Expect them to change.

‼️ A PHP vulnerability is making the rounds, and the honest answer is that **there is no easy exploit for it**. That is not the same thing as saying it is harmless. A memory bug in PHP is a flaw in the interpreter itself, the program that reads and runs every PHP page on a server. When the flaw lets an attacker write past the end of a buffer, the program writes data into memory it does not own. Whether that turns into a takeover depends on what the attacker can place there, and that is a chain of conditions, not a single switch. The article walks through why the obvious paths do not line up: the reachable code paths are narrow, and the values an attacker would need are not the ones the bug hands them. So the practical reading is that patching is still the move, and panicking is not. A bug that resists exploitation today is a bug someone will keep poking at tomorrow, and the version you are running is the only thing you control. Upgrade PHP on the schedule you already have, and treat "no public exploit" as a status, not a guarantee. Early report: these facts come from first disclosures and are not confirmed yet. Expect them to change.

‼️ ‼ Running PHP on Lambda or Elastic Beanstalk? Two flaws just landed in AWS's PHP runtimes, and no fixed version is announced. The problem: AWS published two advisories, CVE-2026-109618 and CVE-2026-109619, affecting the PHP runtimes on its platform. A runtime is the layer that executes your code: you write PHP, AWS supplies the interpreter and the environment. A flaw in that layer is not patched in your repo, it is patched at AWS. How I read it: the advisory is an AWS security bulletin, not a framework vendor's release note. It describes neither the attack vector, nor the affected versions, nor a fix date. That is the usual shape of AWS bulletins: they announce the flaw exists before publishing the analysis. What I take from it: if you deploy PHP on either platform, you cannot patch it yourself today. The only move available is to check what your functions expose, and shrink the surface until the fix ships. What I recommend: audit the public entry points of your PHP functions (URLs, API Gateway triggers, event handlers), put rate limiting or authentication in front of the ones that have none, and watch invocation logs for unusual calls. The bulletin is linked in the reply. Early report: these facts come from first disclosures and are not confirmed yet. Expect them to change.

‼️ ‼ Oracle published a security advisory for CVE-2026-91403, and the fix is out. The problem: a vulnerability in an Oracle product, disclosed through an official security advisory. A CVE is a unique ID assigned to every known flaw, like a case number for a specific incident. Oracle publishes these on a monthly cadence that operations teams know well. How to read it: an advisory with no public technical detail does not say whether the flaw is being exploited in the wild. Oracle also does not state the attack vector or the affected versions in the advisory title. What is certain is that the fix exists, dated October 21, 2026. What follows: an unpatched Oracle CVE on an exposed server stays an open door until the patch lands. The gap between the advisory and the deployment is the only lever the operations team controls. What to do: check whether the affected Oracle product runs in your estate, apply the fix as soon as you can, and watch access logs on exposed services in the meantime. Early report: these facts come from first disclosures and are not confirmed yet. Expect them to change.

Your assistant can answer questions. But can it keep working while the app is closed? nanoMuse is an open-source personal agent for every device you own. > One agent with a name and a look of its own > Does things instead of answering questions > Keeps working while the app is closed > Stops to ask before anything you could not undo The whole set is in one repository under GPL-3.0-or-later, and its README reports 1,000 stars eighteen days after the first release. ⭐ 2.2k stars in 29 days · Swift · GPL-3.0