Claude Code’s “Don’t Ask Again” File Exposes Passwords and Keys in 1,122 Public Copies on GitHub

, , , ,

Anthropic’s AI coding assistant Claude Code saves approved shell commands, including embedded passwords and API keys, in a project’s .claude/settings.local.json file, which has been found exposed in 1,122 public GitHub repositories. Security researchers from Apiiro identified 1,379 different secrets in these files, including database passwords, SSH logins, and cloud service keys, often remaining accessible in git history and forks for over 200 days. Developers are advised to rotate all exposed credentials, remove the files from version control, and avoid embedding secrets directly in commands to prevent such leaks.

https://thehackernews.com/2026/11/claude-codes-dont-ask-again-file.html

Microsoft Paint RCE Fix Now Available, but Windows Update Won’t Deliver It

, ,

Microsoft has released a fix for CVE-2026-83844, a remote code execution vulnerability in Windows Paint, available exclusively through a Microsoft Store update rather than Windows Update. The patched version is 11.2609.401.0, but many systems may still run older versions, requiring administrators to verify and update each user’s Paint installation via PowerShell commands. Microsoft rates the flaw as important with a CVSS score of 8.1, noting exploitation requires specialized knowledge, and no exploitation has been reported to date.

https://thehackernews.com/2026/11/microsoft-paint-rce-fix-now-available.html

So My ISP Was Hijacking My DNS

, , ,

Karthikey Hegde discovered his ISP was hijacking his DNS queries by returning tampered IP addresses, which he verified by comparing plaintext DNS responses with encrypted DNS over HTTPS (DoH). He explains that while running a local DNS resolver like Unbound provides some control, traditional DNS queries remain unencrypted and vulnerable to ISP tampering unless forwarded over encrypted channels like DNS over TLS (DoT) or DoH. He also clarifies that DNSSEC verifies authenticity but does not encrypt traffic, so ISPs can still observe or manipulate DNS requests unless encrypted transport or VPNs are used.

https://www.karthihegde.dev/posts/dns_tampering

Bug Report: Privacy Violation in User-Agent Header · Issue #107757 · Anthropics/claude-Code

, , ,

A security bug report in the anthropics/claude-code repository highlights that the software disclosed users’ email addresses in the User-Agent HTTP header during web searches without consent, leading to privacy violations and increased scam emails. Users noted this behavior persists despite the introduction of a built-in privacy plugin in recent versions; concerns remain about the inclusion of identifiable session data in requests generated by Claude. Contributors recommend removing personal information from User-Agent headers by default to prevent unintentional data exposure to third parties.

https://github.com/anthropics/claude-code/issues/107757

Do the Math: How We Took Over ChatGPT Accounts

, , ,

In October 2026, researchers discovered a critical vulnerability in ChatGPT’s public share feature that allowed attacker-controlled JavaScript to run inside a victim’s authenticated session simply by opening a shared link. By exploiting an unverified postMessage listener in MathLive and bypassing Content Security Policy via YouTube’s JSONP endpoint, the attack executed unsandboxed scripts with full chatgpt.com privileges, enabling enumeration and exfiltration of private conversations and complete account takeover. OpenAI fixed the flaw on October 26 by blocking unauthorized cross-frame messages and updating frontend protections, highlighting the importance of end-to-end verification in sandboxed environments and cross-feature security.

https://www.parameter.ai/labs/we-found-a-gap-at-openai

ChatGPT Lockdown Mode Bypass: Phishing, Malware Delivery, and Silent Data Exfiltration

, ,

Check Point Research discovered a vulnerability in ChatGPT’s recently introduced Lockdown Mode that allowed attackers to bypass its core protections, enabling silent exfiltration of chat data, phishing via spoofed application windows, and malware delivery. The flaw exploited permissive sandbox iframe permissions to open pop-ups outside the sandbox, facilitating covert data theft and social engineering attacks, including zero-click exploitation in the ChatGPT Classic app. OpenAI fixed the issue by removing risky permissions from the HTML-preview sandbox iframe, closing the attack vector.

https://research.checkpoint.com/2026/chatgpt-lockdown-mode-bypass-phishing-malware-delivery-and-silent-data-exfiltration/

Researchers Uncover Prompt Adware Hidden in OpenAI Codex Jailbreak Toolkit

, ,

Cybersecurity researchers from Check Point Research have revealed a prompt adware named Haiou embedded within an OpenAI Codex jailbreak toolkit that modifies the AI agent’s instruction set to inject persistent advertisements for a third-party AI gateway. This prompt adware bypasses traditional payload detection by altering natural language prompts and global configurations, potentially facilitating supply chain attacks and undermining trust in AI agent behavior, especially as such agents grow more integrated into enterprise environments.

https://thehackernews.com/2026/11/researchers-uncover-prompt-adware.html

Google Calendar Phishing Campaign Hijacks Sessions, Bypasses MFA

, , ,

Since August 2026, a phishing campaign dubbed Operation CloudFire has targeted corporate Google users by sending Google Calendar invites that lead to a live reverse proxy capturing real-time Google sessions, effectively bypassing multi-factor authentication. The attackers used multiple layers, including Google redirectors and Cloudflare challenges, to evade detection and leveraged compromised sender addresses to enhance credibility, while Material Security’s team tracked and disrupted the infrastructure by taking down hundreds of domains. Although the campaign appears to have paused, analysts warn the operator has prepared variants for Microsoft 365, signaling potential future threats using similar tactics.

https://www.bleepingcomputer.com/news/security/google-calendar-phishing-campaign-hijacks-sessions-bypasses-mfa/

Hackers Use N-Day Exploits in Zero-Click Attacks on iOS, Android Devices

, , , ,

APT29, a Russian threat group, has conducted a large-scale watering hole campaign targeting iOS and Android users through compromised legitimate and fake websites to deploy platform-specific n-day zero-click exploits. These exploits, including several zero-days patched by Apple and Google, allowed attackers to steal cookies, messaging app databases, credentials, and other sensitive information without user interaction. Security researchers recommend users apply the latest security updates to protect their devices from these ongoing attacks.

https://www.bleepingcomputer.com/news/security/hackers-use-n-day-exploits-in-zero-click-attacks-on-ios-android-devices/

Google Issues Patches for Critical Android Flaw That Could Allow RCE Without User Interaction

, ,

Google released patches in its November 2026 Android security bulletin fixing a critical vulnerability (CVE-2026-58868) affecting Android 14 through 17 that could allow remote code execution without user interaction via a heap buffer overflow. The update also addresses other critical flaws, including a remote denial-of-service issue in Android’s SMS handling, urging users to install the November security patch (2026-11-01 or later) as soon as possible, though no active exploitations have been reported so far.

https://thehackernews.com/2026/11/google-patches-critical-android-flaw.html

CWP2Shell: The Login Page That Never Says No

, ,

VulnCheck disclosed ten critical vulnerabilities in Control Web Panel (CWP), including an unauthenticated remote code execution (CVE-2026-73695) via a module path traversal combined with a missing exit() after a login redirect that allows execution of attacker-controlled PHP modules. This initial foothold lets attackers escalate to root through unauthenticated command injection and SQL injection flaws in the loopback user API, enabling full root access on affected servers; all issues are fixed in CWP 1.29 (October 31, 2026). Given tens of thousands of internet-facing installations expose the vulnerable user panel, operators must upgrade immediately to mitigate high-severity exploit paths actively demonstrated with public exploit tooling.

https://www.vulncheck.com/blog/control-web-panel-pre-auth-module-dispatch-rce

What Exploited Vulnerabilities Have in Common: a Data-Driven Analysis of Real-World Attacks

,

Analysis of approximately 383,000 disclosed software vulnerabilities shows that fewer than 0.5% have been exploited in real-world attacks, with threat actors primarily targeting flaws that enable system takeover such as remote code execution, privilege escalation, and authentication bypass. Exploitation occurs rapidly—often within days of disclosure—while patching critical systems can take weeks to months, underscoring the need for defenders to prioritize remediation based on the potential impact and exposure of vulnerabilities, focusing especially on those that allow full control over highly exposed systems. Microsoft’s security tools assist defenders by continuously discovering internet-facing assets, prioritizing vulnerabilities using exposure and exploitability data, and analyzing attack paths to prevent compromise from spreading within environments.

https://www.microsoft.com/en-us/security/blog/2026/11/02/exploited-vulnerabilities-common-data-driven-analysis-real-world-attacks/

Prompt Adware: a New AI Security Risk

, , ,

Check Point Research identified “prompt adware,” a new security risk where trusted AI agents’ natural-language instructions are persistently modified to influence their behavior, demonstrated by the Haiou jailbreak bundle that injects advertisements into OpenAI Codex responses. This technique, which manipulates AI prompts and configurations rather than executable code, raises concerns about visibility and control over third-party AI components, potentially enabling covert manipulation or data exfiltration in enterprise environments. Organizations are advised to enhance oversight of AI agents, their prompts, and third-party tools to manage these emerging risks as AI adoption grows.

https://blog.checkpoint.com/security/prompt-adware-how-hidden-ai-instructions-could-create-a-new-enterprise-security-risk/

UEFI Bootkit Infected Fully Patched Windows Machines. Is Microsoft to Blame?

, , ,

The UEFI bootkit BlackX has infected fully patched Windows machines by exploiting a long-standing vulnerability in the Windows boot manager that Microsoft patched in 2022 but has not fully neutralized due to delayed revocation of trust in older vulnerable bootloader certificates. Despite multiple updates, Microsoft requires administrators to manually revoke these certificates—a step often overlooked—which allowed BlackX to bypass Secure Boot protections and run malicious code early in the boot process. This incident highlights challenges in Microsoft’s patch deployment and communication, leaving systems exposed to persistent and stealthy boot-level malware.

https://arstechnica.com/security/2026/11/uefi-bootkit-infected-fully-patched-windows-machines-is-microsoft-to-blame/

PHP2Shell: What We Know About the Threat Two Weeks Later

, ,

Two critical PHP vulnerabilities known as PHP2Shell, disclosed in late October 2026, could allow remote code execution on about 70% of websites globally, yet mass exploitation reports remain limited. While emergency patches have been released alongside a subsequent fix for a related Brotli library flaw, many servers remain unpatched, and successful attacks—such as those targeting Palo Alto Networks PAN-OS devices—have demonstrated real-world impact. Experts advise urgent patching combined with thorough system audits and using specialized detection tools to identify potential compromises.

https://www.kaspersky.com/blog/php2shell-two-weeks-after-patches/56647/

Scroll to Top