Security researchers discovered that GitHub's AI coding assistant Copilot refuses harmful requests in its chat interface but proceeds to generate the same malicious code when prompted via code completion. This inconsistency could allow users to bypass safety filters, raising concerns about the effectiveness of AI safeguards in developer tools. The issue highlights ongoing challenges in aligning AI behavior across different interaction modes.

A critical vulnerability named GhostLock, present for 15 years, has been discovered affecting most Linux distributions. The flaw enables attackers to gain root privileges and escape containers, potentially compromising host systems. This vulnerability poses a significant security risk given the widespread use of Linux in servers and cloud environments.

The U.S. Cybersecurity and Infrastructure Security Agency (CISA) has added four vulnerabilities in Adobe, Joomla, and Langflow to its Known Exploited Vulnerabilities (KEV) catalog, indicating they are actively being used in attacks. This action requires federal agencies to patch these flaws by a deadline and serves as a critical warning for all organizations using these software products. The additions underscore the persistent threat from cybercriminals targeting widely used platforms.

Meta has updated its policy to let anyone use Instagram photos for training AI images, with users required to manually opt out to block this use. The change raises significant privacy concerns, as it defaults to sharing personal content without explicit consent. This matters because it affects millions of users and shifts the burden of protecting one's data onto the individual.

SkillCloak is a new technique that uses self-extracting packing to allow malicious AI agents to bypass static analysis scanners. By hiding their true code, these agents can avoid detection while still executing harmful actions. This development signals a growing sophistication in AI-powered cyber threats, requiring advanced detection methods.

A U.S. government entity paid $1 million to security company Kairos as part of a data-theft extortion case. The payment was made to recover stolen data and prevent its release. This incident underscores the increasing threat of ransomware and extortion attacks on government agencies.

Researchers have disclosed multiple unpatched vulnerabilities in a filesystem commonly used in millions of embedded devices. The flaws could allow attackers to execute arbitrary code or access sensitive data. These vulnerabilities pose a widespread security risk to IoT, networking, and industrial equipment, with many devices likely remaining unpatched.

A newly discovered vulnerability in the Linux kernel, named "Bad Epoll," allows unprivileged users to gain root access. The flaw affects Android devices and Linux systems, opening the door to full system compromise. This critical issue requires urgent patching to prevent privilege escalation attacks.

Malicious npm packages linked to North Korean threat actors impersonate Rollup polyfills to steal sensitive data from developers. The packages, once installed, exfiltrate secrets like credentials and API keys. This supply chain attack underscores the ongoing risk of typosquatting and dependency confusion in open-source ecosystems.