I came across a project, “0day Rubbish”, that states it will continuously disclose 0-day vulnerabilities discovered by an AI-driven research process (a multi-LLM ensemble of Claude, OpenAI, DeepSeek and GLM). For each vulnerability the project publishes a full technical analysis together with a working exploit script and step-by-step reproduction instructions. The project’s content is hosted at:

https://0day-rubbish.com/blog

The first batch already covers ten previously undisclosed vulnerabilities across eight enterprise products, ranging from CVSS 7.3 to 9.8. The list below summarises them for the community’s awareness.

The vulnerabilities, in descending CVSS order:

  1. InterSystems IRIS - FolderManager Property Injection RCE

  2. AdRem NetCrunch - Cross-Session Hijack RCE

  3. Altus BluePlant - Hardcoded Credentials RCE

  4. Brekeke SIP Server - Unauthenticated Nashorn JS Engine RCE

  5. Brekeke SIP Server - Unauthenticated Zip Slip Webshell RCE

    • Product: Brekeke SIP Server
    • Version: v3.19.1.8p1 (Evaluation Edition); versions sharing the ProvisioningModelImport / Zip.extractAll code
    • CVSS: 9.8 (AV:N/AC:L/PR:N/UI:N/S:C/C:H/I:H/A:H)
    • Class: Unauthenticated remote code execution (Zip Slip)
    • Link: https://0day-rubbish.com/blog/brekeke-sip-server-zipslip-rce

  6. DataSunrise Suite - Email Verification Code Brute Force RCE

  7. Brekeke SIP Server - Authentication Fail-Open Primitive

  8. Acumatica ERP - Customization Publish Webshell RCE

    • Product: Acumatica ERP (.NET ERP platform)
    • Version: 2026 R1 (build 26.101.0225); versions sharing the CstRollbackList.WriteFile / CstFileWriter.UpdateFile publish path
    • CVSS: 8.8 (AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H)
    • Class: Authenticated remote code execution (webshell via path traversal in customization publish)
    • Link: https://0day-rubbish.com/blog/acumatica-customization-webshell-rce

  9. AdRem NetCrunch - Startup Script RCE

  10. Altus iX Developer - XAML Deserialization RCE

    • Product: Altus iX Developer (SCADA HMI engineering IDE)
    • Version: 2.53.65422; versions sharing the ScreenXamlSerializer / XamlReader.Load path
    • CVSS: 7.3 (AV:L/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H)
    • Class: Local / user-interaction remote code execution via malicious screen file
    • Link: https://0day-rubbish.com/blog/altus-ix-developer-xaml-rce

Observations on the project’s disclosure posture

The project presents its work as defensive security research and, according to its own pages, operates a risk-driven disclosure model: it publishes full root-cause details, exploit code and reproduction steps so that vendors can verify and patch and defenders can detect and mitigate. Each advisory page includes a self-contained exploit script (Python or shell, pure standard library where possible) against a default or evaluation deployment of the affected product; the unauthenticated entries require no credentials beyond product defaults. The project attributes discovery to its automated AI research process. I am noting this only to characterise what the project says about itself; I make no claim about its disclosure ethics.

Reference

https://0day-rubbish.com/blog

Sent through the Full Disclosure mailing list https://nmap.org/mailman/listinfo/fulldisclosure Web Archives & RSS: https://seclists.org/fulldisclosure/

  • KnilAdlez [none/use name]@hexbear.netEnglish
    14·
    4 months ago

    I cannot find anyone behind this, though Claude is part of some of the commits (just edits to the README though). I’m going to guess that this website, github, and the code that calls the AI models are all done by an AI. I suspect this is another stunt, like the Claude ‘model too powerful for public release’ and chatgpt ‘the model escaped containment and started attacking huggingface’. A point against this would be that it does claim to be using deepseek (I would expect such a stunt to avoid Chinese models) but I think it is still a strong possibility

    • seas_surround [he/him]@hexbear.netEnglish
      6·
      4 months ago

      Sifting through oceans of bad code no human has ever cared about and identifying known patterns from its training data is one of the few cases where using LLMs actually makes sense. I don’t know about the claims of this repo specifically, but someone was bound to do something like this eventually. There is such a massive quantity of poorly written software out there which only snuck beneath notice by volume and obscurity, it’s very plausible all the examples here are such low hanging fruit

      • KnilAdlez [none/use name]@hexbear.netEnglish
        4·
        4 months ago

        You’re not wrong and there are several AI bots that do that and post patches as git pulls. It’s the way everything is AI generated, none of the links at the bottom of the page that could identify whoever made this work, and the WHOIS data is hidden through a proxy, reeks of hidden motive. Right after the ChatGPT ‘leak’ and OpenAI and Anthropic lobbying for protections on open source models, I think this is just another “ooh the AI models are too scawy :c” publicity stunt.

    • VOLCEL_POLICE [it/its]@hexbear.netBEnglish
      4·
      4 months ago

      The VOLCEL POLICE are on the scene! PLEASE KEEP YOUR VITAL ESSENCES TO YOURSELVES AT ALL TIMES.

      نحن شرطة VolCel.بناءا على تعليمات الهيئة لترويج لألعاب الفيديو و النهي عن الجنس نرجوا الإبتعاد عن أي أفكار جنسية و الحفاظ على حيواناتكم المنويَّة حتى يوم الحساب. اتقوا الله، إنك لا تراه لكنه يراك.

      volcel-police