loophole

Security expert Sean Heelan published a blog post on May 22 reporting that he had discovered a major zero-day vulnerability in the Linux kernel with the help of OpenAI's o3 inference model, tracked as CVE-2025-37899. Heelan said that he initially intended to test OpenAI's o3 inference model through a code audit, but instead discovered that the AI was able to The complex "use-after-free" vulnerability, which is autonomously recognized by AI, exists in the Linux kernel's SMB protocol implementation...

In a May 27 post, security firm Legit Security disclosed a prompt injection vulnerability in GitLab's AI assistant GitLab Duo, which allows a hacker to make GitLab Duo output whatever the hacker wants by embedding a prompt (prompt injection). GitLab Duo, an AI assistant built on Claude, a large language model developed by Anthropic, was released on June 6, 2023, according to ...

March 7, 2011 - Anthropic's latest coding tool, Claude Code, is experiencing some technical issues. Users on GitHub have reported a vulnerability in the tool's auto-update feature that has caused some workstations to become unstable or even not function properly. According to 1AI, the root of the problem is that Claude Code's auto-update command contains an error code. When the tool is installed with "root" or "superuser" privileges, these error commands allow the application to modify file directories that are normally restricted...

OpenAI announced the introduction of Memory for ChatGPT in February this year, and in early September this year, the feature was officially opened to all users, the main purpose of the feature is to "set presets", which will help the AI to the user to better meet the needs of the answer. However, researcher Johann Rehberger has now reported that there is a vulnerability in the PC version of the ChatGPT client that could allow hackers to gain access to a user's conversation history. Johann Rehberger calls the vulnerability...

According to NewAtlas, researchers have successfully hacked more than half of the sites tested using a team of autonomous, collaborative GPT-4 bots that can coordinate their actions and generate new "helpers" as needed. What's even more surprising is that they utilized previously unknown, real-world zero-day vulnerabilities. Pexels A few months ago, the same researchers published a paper claiming that they were able to use GPT-4 to automatically exploit "N da...