AI Security News: November 17, 2026
Fully automated. Found, summarized and published by an automated AI pipeline. How posts are labeled
1. Google Threat Intelligence Group Reports Attackers Running Fully Automated Vulnerability Discovery and Exploitation
Google Threat Intelligence Group and Mandiant published an AI Threat Tracker update saying adversaries have moved from basic prompting to autonomous multi-agent offensive frameworks, naming ARTEX, CAIRN, CyberStrikeAI and PentAGI. The report describes LLM-driven differential analysis of patches to weaponize n-day vulnerabilities, LLMjacking where stolen LLM credentials are resold through underground proxies, and espionage groups manipulating coding assistant configurations to drop backdoors. The Decoder reports GTIG found an exposed server running an exploit_agent framework that finds, exploits and runs attacks without a human in the loop, which Google rates at Level 5 of its scale. Source
2. FBI Flash Advisory Says the Open-Source Artex Tool Compromised Networks in Under Three Hours
The FBI issued a flash advisory warning that attackers are using autonomous AI penetration testing tools, led by the open-source Artex, which can be connected to hosted frontier models, open-source models and non-U.S. models. According to TechCrunch, Artex-linked intrusions can go from targeting to compromise in under three hours and have been tied to the theft of data on at least 68,000 people from South Korean financial institutions and to the compromise of Microsoft 365 environments of at least 2,250 ConnectWise customers. The advisory says intrusions are noisy and usually rely on simple weaknesses such as missing multi-factor authentication and exposed credentials. Source
3. Anthropic Cuts Its Vulnerability Disclosure Window to 30 Days for High- and Critical-Severity Bugs
Anthropic shortened its coordinated vulnerability disclosure window from 90 days to 30 days for high- and critical-severity issues and to 45 days for all others, effective for reports made on or after January 1, 2027. The post says frontier models are finding vulnerabilities well enough that rediscovery is increasingly likely and that fixes can now move faster. Anthropic will also give maintainers who are past the old 90-day deadline an extra 30 days before publishing those reports on its dashboard. Source
4. Wiz Launches an AI Security Agent That Maps Application-Layer Agent Flaws to Cloud Attack Paths
Wiz introduced the Wiz AI Security Agent, which discovers AI agents, including custom agents defined only in code, and identifies their models, tools, skills, MCP servers, gateways and guardrails. It analyzes agent code for AI-specific flaws such as prompt injection leading to arbitrary command execution, then feeds findings into the Wiz Security Graph to check external exposure, internet egress, identities and permissions. Wiz gives a hypothetical BigQuery agent as an example, where a code flaw plus broad database permissions forms a complete attack path. Source
5. GitHub Lets Organizations Start AI Scan on Up to 20 Repositories at Once, in Public Preview
GitHub added bulk start for AI Scan for repositories, which finds security vulnerabilities on a repository’s default branch, including in languages CodeQL does not support and additional issues in languages it does. Organizations can start scans for up to 20 repositories at once when enabling the organization-level setting or from the code scanning organization alerts page. Findings appear in the code scanning backlog alongside CodeQL alerts, and a Copilot license is not required. Source