<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
<channel>
  <title>HanulSec AI Security News</title>
  <link>https://hanulsec.com/news/</link>
  <atom:link href="https://hanulsec.com/news/feed.xml" rel="self" type="application/rss+xml"/>
  <description>Daily briefs on new AI security vulnerabilities and incidents: what happened, who is affected and what to do now.</description>
  <language>en</language>
  <lastBuildDate>Sat, 21 Nov 2026 12:00:00 GMT</lastBuildDate>
  <item>
    <title>An AI server-admin tool could run any command on your servers.</title>
    <link>https://hanulsec.com/news/mcp-ssh-manager-command-injection/</link>
    <guid isPermaLink="true">https://hanulsec.com/news/mcp-ssh-manager-command-injection/</guid>
    <pubDate>Sat, 21 Nov 2026 09:00:00 GMT</pubDate>
    <category>mcp-ssh-manager · CVE-2026-129995</category>
    <description>A command injection in mcp-ssh-manager’s backup tools let an AI agent run any command on managed servers. CVSS 8.8. Update to 3.8.5 now.</description>
    <enclosure url="https://hanulsec.com/social/hanulsec-ai-security-news-mcp-ssh-manager-command-injection.png" type="image/png" length="0"/>
  </item>
  <item>
    <title>An AI document reader could reach files outside its folder.</title>
    <link>https://hanulsec.com/news/aws-document-loader-mcp-path-traversal/</link>
    <guid isPermaLink="true">https://hanulsec.com/news/aws-document-loader-mcp-path-traversal/</guid>
    <pubDate>Sat, 21 Nov 2026 09:00:00 GMT</pubDate>
    <category>AWS Document Loader MCP · CVE-2026-130901</category>
    <description>AWS fixed a path traversal in its Document Loader MCP Server that could let AI assistants read files outside their folder in CI setups. Update to 1.0.21.</description>
    <enclosure url="https://hanulsec.com/social/hanulsec-ai-security-news-aws-document-loader-mcp-path-traversal.png" type="image/png" length="0"/>
  </item>
  <item>
    <title>An AI gateway flaw is now on CISA’s exploited list.</title>
    <link>https://hanulsec.com/news/litellm-guardrail-sandbox-escape-cisa-kev/</link>
    <guid isPermaLink="true">https://hanulsec.com/news/litellm-guardrail-sandbox-escape-cisa-kev/</guid>
    <pubDate>Fri, 20 Nov 2026 09:00:00 GMT</pubDate>
    <category>LiteLLM · CVE-2026-40217</category>
    <description>CISA lists a LiteLLM guardrail sandbox escape (CVE-2026-40217) as exploited. Who is affected, why AI gateways are targets, and the steps to take now.</description>
    <enclosure url="https://hanulsec.com/social/hanulsec-ai-security-news-litellm-guardrail-sandbox-escape-cisa-kev.png" type="image/png" length="0"/>
  </item>
  <item>
    <title>A malicious MCP server could collect your secrets.</title>
    <link>https://hanulsec.com/news/mcp-python-sdk-oauth-secret-leak/</link>
    <guid isPermaLink="true">https://hanulsec.com/news/mcp-python-sdk-oauth-secret-leak/</guid>
    <pubDate>Fri, 20 Nov 2026 09:00:00 GMT</pubDate>
    <category>MCP Python SDK · CVE-2026-115563</category>
    <description>A flaw in the official MCP Python SDK let a hostile MCP server collect OAuth secrets. Who is affected, the fixed versions and the extra steps to take.</description>
    <enclosure url="https://hanulsec.com/social/hanulsec-ai-security-news-mcp-python-sdk-oauth-secret-leak.png" type="image/png" length="0"/>
  </item>
  <item>
    <title>A research paper hid orders for AI reviewers.</title>
    <link>https://hanulsec.com/news/hidden-prompt-injection-journal-paper/</link>
    <guid isPermaLink="true">https://hanulsec.com/news/hidden-prompt-injection-journal-paper/</guid>
    <pubDate>Fri, 20 Nov 2026 09:00:00 GMT</pubDate>
    <category>Prompt injection · Peer review</category>
    <description>A published paper hid instructions telling AI reviewers to write a positive review. Why this prompt injection matters for any business AI that reads documents.</description>
    <enclosure url="https://hanulsec.com/social/hanulsec-ai-security-news-hidden-prompt-injection-journal-paper.png" type="image/png" length="0"/>
  </item>
  <item>
    <title>An AI agent turned a security patch into an exploit on its own.</title>
    <link>https://hanulsec.com/news/google-ai-agent-patch-to-exploit/</link>
    <guid isPermaLink="true">https://hanulsec.com/news/google-ai-agent-patch-to-exploit/</guid>
    <pubDate>Fri, 20 Nov 2026 09:00:00 GMT</pubDate>
    <category>Google Threat Intelligence</category>
    <description>Google’s threat team found an attacker’s AI agent that turned security patches into working exploits with no human help. What it means for patching.</description>
    <enclosure url="https://hanulsec.com/social/hanulsec-ai-security-news-google-ai-agent-patch-to-exploit.png" type="image/png" length="0"/>
  </item>
  <item>
    <title>A “read-only” AI tool could write files anywhere.</title>
    <link>https://hanulsec.com/news/aws-pricing-mcp-server-file-write/</link>
    <guid isPermaLink="true">https://hanulsec.com/news/aws-pricing-mcp-server-file-write/</guid>
    <pubDate>Fri, 20 Nov 2026 09:00:00 GMT</pubDate>
    <category>AWS Pricing MCP Server · CVE-2026-126816</category>
    <description>CVE-2026-126816 let the AWS Pricing MCP Server write files outside its folder, and it was marked read-only so clients could auto-approve it. What to do now.</description>
    <enclosure url="https://hanulsec.com/social/hanulsec-ai-security-news-aws-pricing-mcp-server-file-write.png" type="image/png" length="0"/>
  </item>
  <item>
    <title>An MCP gateway could be steered into your internal network.</title>
    <link>https://hanulsec.com/news/aws-mcp-gateway-registry-ssrf/</link>
    <guid isPermaLink="true">https://hanulsec.com/news/aws-mcp-gateway-registry-ssrf/</guid>
    <pubDate>Fri, 20 Nov 2026 09:00:00 GMT</pubDate>
    <category>mcp-gateway-registry · CVE-2026-125926</category>
    <description>An SSRF flaw in the open-source mcp-gateway-registry let users steer the gateway into internal networks and cloud credentials. Fixed in 1.35.0.</description>
    <enclosure url="https://hanulsec.com/social/hanulsec-ai-security-news-aws-mcp-gateway-registry-ssrf.png" type="image/png" length="0"/>
  </item>
  <item>
    <title>A breach at an AI vendor exposed customers’ chats.</title>
    <link>https://hanulsec.com/news/fireworks-ai-breach-notion-chat-data/</link>
    <guid isPermaLink="true">https://hanulsec.com/news/fireworks-ai-breach-notion-chat-data/</guid>
    <pubDate>Fri, 20 Nov 2026 09:00:00 GMT</pubDate>
    <category>Fireworks AI · Notion</category>
    <description>A breach at AI provider Fireworks exposed AI chat transcripts from Notion. What happened, who is affected and how to manage AI vendor risk.</description>
    <enclosure url="https://hanulsec.com/social/hanulsec-ai-security-news-fireworks-ai-breach-notion-chat-data.png" type="image/png" length="0"/>
  </item>
  <item>
    <title>Cloned executive voices, real wire transfers.</title>
    <link>https://hanulsec.com/news/ai-voice-clone-ceo-fraud-129m-indictment/</link>
    <guid isPermaLink="true">https://hanulsec.com/news/ai-voice-clone-ceo-fraud-129m-indictment/</guid>
    <pubDate>Fri, 20 Nov 2026 09:00:00 GMT</pubDate>
    <category>Voice cloning · Fraud</category>
    <description>Three men were charged in New York over an alleged scheme using AI-cloned executive voices to trigger wire transfers. How to protect your payments.</description>
    <enclosure url="https://hanulsec.com/social/hanulsec-ai-security-news-ai-voice-clone-ceo-fraud-129m-indictment.png" type="image/png" length="0"/>
  </item>
  <item>
    <title>A security setting is only as strong as the app around it.</title>
    <link>https://hanulsec.com/news/chatgpt-lockdown-mode-sandbox-escape/</link>
    <guid isPermaLink="true">https://hanulsec.com/news/chatgpt-lockdown-mode-sandbox-escape/</guid>
    <pubDate>Fri, 20 Nov 2026 09:00:00 GMT</pubDate>
    <category>ChatGPT · Check Point Research</category>
    <description>Check Point found a sandbox escape in ChatGPT HTML previews that got around Lockdown Mode. OpenAI fixed it. What it means for apps that render AI output.</description>
    <enclosure url="https://hanulsec.com/social/hanulsec-ai-security-news-chatgpt-lockdown-mode-sandbox-escape.png" type="image/png" length="0"/>
  </item>
</channel>
</rss>
