Tracking Threats, Incidents, and Vulnerabilities
Tracking Threats, Incidents, and Vulnerabilities
Amazon is experiencing a trend of outages, some linked to AI coding tools, prompting a meeting with engineers to address the issue. The company will require a senior engineer's sign-off for AI-assisted changes and focus on improving website availability. AWS also experienced incidents involving AI coding assistants, including a 13-hour interruption of a cost calculator.
Anthropic identified 22 vulnerabilities in Firefox using its AI model, Claude Opus 4.6. Among these, 14 are high severity, discovering a significant number of issues addressed in Firefox 148. The model's efficiency in finding issues, compared to creating exploits, raises security concerns, highlighting AI's role in enhancing browser security. Mozilla reported additional vulnerabilities found through this collaboration, showcasing the benefits of AI-assisted analysis for continuous improvement in security.
https://thehackernews.com/2026/03/anthropic-finds-22-firefox.html
APT36, known as Transparent Tribe, shifts from conventional malware to “vibeware,” an AI-generated model producing numerous low-quality implants using niche languages like Nim, Zig, and Crystal. This evolution aims to evade detection and employs trusted cloud services for command and control. Despite technical flaws leading to ineffective malware, this model's production volume overwhelms defenses, indicating a trend towards automated, high-volume cyberattacks. Their targeted attacks focus on the Indian government, utilizing sophisticated social engineering tactics and established frameworks alongside new, poorly coded variants. Overall, APT36 embraces a strategy of integrating AI into malware design, resulting in mass-produced threats lacking true innovation but full of operational risk.
https://businessinsights.bitdefender.com/apt36-nightmare-vibeware
Large language models (LLMs) can accurately unmask pseudonymous users on social media platforms, thereby undermining the privacy afforded by pseudonymity. Researchers found that LLMs can achieve high recall and precision rates in identifying users based on their online activity, posing risks of doxxing, stalking, and targeted advertising. The study highlights the need for stronger privacy protections and suggests mitigations, such as rate limits on data access and monitoring for LLM misuse.
IDPI exploits hidden instructions in web content processed by LLMs, causing unauthorized actions without direct interaction. Recent evidence shows substantial real-world malicious exploitation, including AI ad review evasion and SEO manipulation targeting phishing. 22 techniques were identified, necessitating proactive defenses against such threats. Understanding and mitigating web-based IDPI is crucial for the safety of AI systems integrated into web operations.
https://unit42.paloaltonetworks.com/ai-agent-prompt-injection/
A user demonstrated AI's vulnerability by fabricating an article about tech journalists eating hot dogs, which major chatbots mistakenly accepted as fact within hours. This highlights AI's unreliability and potential for misinformation, especially as these systems gain wider trust.
https://www.schneier.com/blog/archives/2026/02/poisoning-ai-training-data.html
TLDR: Research reveals nearly 3,000 Google Cloud API keys publicly exposed, allowing unauthorized access to sensitive Gemini endpoints post-enablement. Users advised to secure their projects and rotate keys. Google is addressing the issue.
https://thehackernews.com/2026/02/thousands-of-public-google-cloud-api.html
A hacker exploited Anthropic’s AI chatbot, Claude, to breach Mexican government agencies, stealing 150 gigabytes of sensitive data, including taxpayer and voter records. The hacker used Claude to identify vulnerabilities, write scripts, and automate data theft, bypassing Claude’s guardrails by posing as a bug bounty hunter. The attack highlights the growing trend of cybercriminals using AI tools to enhance their hacking capabilities.
Check Point Research identified critical vulnerabilities in Anthropic’s Claude Code enabling remote code execution and API key theft through malicious project configurations. Attackers can exploit Hooks and Model Context Protocol to execute unauthorized commands and intercept API communications. All discovered vulnerabilities have been remediated by Anthropic. Developers must carefully scrutinize project configurations to prevent configuration-based attacks, treating them with the same caution as executable code.
A new variant of Atomic macOS Stealer (AMOS) is being distributed through malicious OpenClaw skills, exploiting AI agentic workflows to trick users into installing the malware. The malware, disguised as a harmless skill, uses a fake dialogue box to request the user’s password and then exfiltrates sensitive data, including Apple and KeePass keychains, user documents, and system information. TrendAI™ Managed Detection and Response (MDR) customers are protected from this threat.
Hillai Ben Sasson and Dan Segev, researchers at Wiz, discovered vulnerabilities in every major AI platform they targeted over two years of research. Their findings, to be presented at the RSAC Conference, highlight the importance of focusing on AI infrastructure security across model training, inference, application, and cloud layers. The researchers emphasize the need for regular security reviews and compliance checks to address the rapidly evolving threat landscape.
https://www.darkreading.com/application-security/lessons-ai-hacking-model-every-layer-risky
Russian-speaking hacker used AI to breach 600 Fortinet firewalls in 55 countries within five weeks, exploiting weak credentials and exposed interfaces without zero-day exploits. The attack involved automating access and reconnaissance tasks with AI-generated tools, leading to stolen configurations and credentials. Recommendations for FortiGate admins include disabling internet exposure of management interfaces and enabling MFA.
User hacked ChatGPT and Google's AI in 20 minutes. Demonstrated that AI tools can easily be manipulated to spread misinformation, even about serious topics. Created a fake ranking of “best tech journalists at eating hot dogs,” and AI accepted it as fact. Experts say AI is now easier to trick, raising concerns about misinformation's impact on public safety. Solutions include enhancing disclaimers and promoting critical thinking when using AI for information.
LLM-generated passwords seem strong but are insecure due to their predictable nature, as LLMs are designed to predict tokens, not create random characters. Users unknowingly use these weak passwords, mistaking them for secure options. Testing reveals that popular LLMs like GPT, Claude, and Gemini generate passwords with predictable patterns and low entropy, risking brute-force attacks. Recommendations include avoiding LLM-generated passwords and prioritizing secure password generation methods in coding contexts.
https://www.irregular.com/publications/vibe-password-generation
Anthropic introduces Claude Code Security, an AI tool for scanning codebases for vulnerabilities and suggesting patches. Initially available to select enterprise users after extensive testing, it aims to automate software security reviews. The tool improves vulnerability detection, with promises of faster discovery and reduced error rates, though experienced human oversight is still recommended for complex issues. Users must apply for access and agree to scan only their own code.
https://cyberscoop.com/anthropic-claude-code-security-automated-security-review/
