AI Security News: November 12, 2026
Fully automated. Found, summarized and published by an automated AI pipeline. How posts are labeled
1. HoneyLabs: A Credential Scanner Added 32 AI Coding Assistant Paths, Including Claude Code’s Token File and MCP Configs, to Its Wordlist
HoneyLabs published an analysis on November 11 of unsolicited traffic to its honeypots between August 27 and November 9. A credential scanner running from Google Cloud (AS396982) that had requested the same 543 paths daily added 49 new ones within a single second on September 4, of which 32 belong to AI coding tools: /.claude/.credentials.json (where Claude Code keeps OAuth tokens on machines without a system keychain), /.claude.json, /.cursor/mcp.json, /.vscode/mcp.json, /.mcp.json, Windsurf, Codeium, Aider and Continue configs, and LiteLLM config files. The remaining 17 were shell history and startup files. The scanner sent 366,845 requests in its first eleven days from 2,894 addresses, 2,499 of them seen on one day only, across 203 ports; HoneyLabs concludes that address blocklists are close to useless and recommends blocking on paths. That scanner requested the AI files on four days and then disappeared, but other clients sent 8,067 further requests for the same files between September 8 and November 9. The post notes that an MCP config’s env block typically carries tokens for other services such as GitHub, a database or Slack, and that the data shows what scanners request, not whether any server answered. Source
2. Arctic Wolf Links Self-Hosted AntiFox Pentesting Agents on Alibaba Cloud to Webshell Deployment and AWS Secret Theft
BleepingComputer reported on November 11 that attackers are hosting self-managed instances of AntiFox, an open-source agentic penetration testing platform that began as a fork of ARTEX 0.1.1, on Alibaba Cloud and using them against web services and AWS environments. Arctic Wolf researchers tied the infrastructure to vulnerability exploitation, webshell deployment, credential validation, configuration discovery and AWS secret harvesting. In one case an ASPX webshell was placed on an internet-exposed IIS server and used to run OS commands, read application settings and database connection strings, and copy configuration and log files to a web-accessible directory. The attacker then used valid IAM credentials and assumed EC2 roles, cycling through identities and making hundreds of API requests in about 30 minutes; most were denied, but some succeeded and returned production secrets, and in a separate incident a valid identity was used to read, copy and delete S3 objects. Arctic Wolf recommends correlating IIS webshell activity with unusual AWS access and reviewing successful calls inside bursts of denials. An Amazon spokesperson told BleepingComputer that AWS services were not compromised and that the credentials came from customers’ compromised applications. AntiFox remains on GitHub at version 2.0; its terms prohibit use against live systems but the article says no technical controls enforce that. Source Arctic Wolf hunting pack
3. Notion Says the Fireworks AI Breach Exposed Notion AI Chat Transcripts Held for Its Insights Feature
Notion disclosed that the breach at its AI infrastructure provider Fireworks AI exposed Notion AI chat transcripts in a small number of workspaces, BleepingComputer reported on November 11. Per the timeline in the report, Fireworks alerted Notion on November 2 about an exposed dataset said to hold only synthetic test data; Notion cut off traffic to Fireworks, deleted its data from the vendor’s systems and halted training runs by November 3; and on November 8 Fireworks confirmed that customer data had been accessed. The transcripts came from the dataset behind Insights, a feature that shows workspace administrators how teams use AI, which is now disabled. Notion states that for the vast majority of impacted workspaces the chats were unattributable, meaning not labeled with an author or workspace, and that transcripts did not contain tool call results; it puts the affected share at under 1% of users and says Fireworks had contractually committed to zero data retention for Notion data. Affected workspace owners can review the data through a security incident review screen, and Notion advises rotating any passwords, API keys or other credentials typed into affected chats. The publication date of Notion’s own advisory was not confirmed. [NEEDS VERIFICATION] Source Notion advisory
4. Cequence Data: A Credential-Stuffing Operation Posing as a Consumer AI Agent Outnumbered Real Agent Visits Across Five Industries
Help Net Security reported on November 12 on Cequence Security data showing abuse in both directions around agent traffic. Over one week in October, an online marketplace received 414,210 visits from traffic carrying the name of a widely used consumer AI agent; every visit went to the sign-in page and every one returned errors, which Cequence attributes to credential stuffing, at about one and a half times the volume of all genuine agent visits it counted across five industries that week. In the other direction, automated traffic at a Tier 1 telecom carrier grew from roughly 100 requests a day to between 19,000 and 32,000 within three weeks of a consumer platform launching a personal agent, signing in, completing multi-factor authentication and adding devices to carts while presenting as a desktop browser; a jump to about 98% adoption of a new browser version within 48 hours exposed it as a managed fleet. In a sample of 281,627 agent visits, 85% only read pages or compared products, and about one in seven went on to booking, buying, signing in or account management. At one consumer brand, behavioral analysis blocked 3.51 million unauthorized login attempts in 31 days, about 82% of which scored too low to trip conventional bot detection. The article does not link the underlying Cequence report. Source