Agent news brief: 9 November 2026 2026-11-09 ยท https://agentnewsdaily.com/brief/2026-11-09 Check Point details cheap criminal campaigns run with ARTEX agents; Apiiro finds Claude Code approvals leaked secrets; Railway gives its account. 1. Check Point says single operators ran criminal campaigns with ARTEX agents for a few hundred dollars Check Point Research says forks of ARTEX, an AI penetration-testing platform from China built for authorised testing, are being used by criminals. It found 1,608 live, publicly reachable ARTEX panels and describes three campaigns, each run by one person and close to fully autonomous once a goal was set. They stole iOS code-signing certificates and more than 190,000 game account credentials, and made a failed attempt to drain cryptocurrency. Check Point estimates each campaign cost between about US$145 and US$1,500 in model use. - Check Point Research: https://research.checkpoint.com/2026/hacking-on-autopilot-inside-artex-cybercrime-campaigns/ 2. Apiiro says Claude Code's 'don't ask again' approvals put secrets in public repositories According to Cybersecurity News, security firm Apiiro found real credentials in 1,122 of 104,185 public copies of Claude Code's local settings file. Choosing 'Yes, and don't ask again' saves the approved command, including any password or key in it, to that file. Counting earlier commits and agent transcripts, Apiiro put the total at 2,539 secrets across 1,994 repositories, and it said common secret scanners missed 215 of the affected files. The article includes no direct response from Anthropic. - Cybersecurity News: https://cybersecuritynews.com/claude-codes-yes-and-dont-ask-again/ 3. Railway gives a fuller account of the engineer's agent that probed Primcloud CSO Online reports that Railway CEO Jake Cooper said the engineer built the testing tool on his own, using personal accounts and his own money, ran it once against one provider and stopped it within 15 minutes. Primcloud founder Josh Manders told the outlet he saw two scans of all 65,536 addresses on Primcloud's internal network, and another access attempt more than 30 minutes after his first post. Primcloud says nothing was accessed. - CSO Online: https://www.csoonline.com/article/4245637/railways-ai-agents-go-off-the-track-on-a-rivals-network.html - Runtime Wire: https://runtimewire.com/article/railway-cofounder-apologizes-after-employee-accused-of-trying-to-breach-primclou 4. Researcher used AI agents to find a decade-old OpenSSH login flaw BleepingComputer reports that researcher Ozgun Kultekin found the flaw with AI agents that ran about 17,600 authentication tests across different OpenSSH configurations. On musl-based systems such as Alpine Linux, a very long password can let attackers log into accounts that are meant to refuse password logins. The fix is in OpenSSH 10.6p2. Systems based on glibc, and those using PAM, are not affected. - BleepingComputer: https://www.bleepingcomputer.com/news/security/openssh-flaw-lets-attackers-log-into-password-disabled-linux-accounts/ 5. Apple reportedly wants Siri to act more like an agent as early as next year 9to5Mac, citing Bloomberg's Mark Gurman, reports that Apple could make Siri 'more like an agent' as early as 2027. The personalised Siri that shipped with iOS 27 still does not take actions on its own initiative. The report suggests Apple's existing App Intents framework could let Siri carry out tasks more cheaply than relying on computer-use abilities. - 9to5Mac: https://9to5mac.com/2026/11/08/apple-working-on-agentic-siri-ai-next-year-report 6. Perplexity launches Teammates, persistent agents for ongoing work Perplexity said on 4 November that Teammates, long-running agents that plan projects, remember past work and send progress updates and approval requests, are available to Max subscribers on the web, with iOS to follow. The company says each one runs in a separate environment with its own browser sessions, and users can take over or redirect a Teammate partway through a task. - Perplexity: https://www.perplexity.ai/hub/blog/teammates-are-agents-for-complex-ongoing-work 7. CrowdStrike adds an AI agent kill switch and agrees to buy Fiddler and Dreadnode CrowdStrike announced at Fal.Con Europe on 3 November a Kill Switch in Falcon Guardian that lets authorised administrators stop an AI agent, block it from relaunching and cut its network access. Every use is logged. It also agreed to acquire agent-monitoring firm Fiddler AI and AI security testing company Dreadnode. CrowdStrike has not disclosed the price; Decrypt cites The Wall Street Journal putting the Fiddler deal at about US$400 million. - The Next Web: https://thenextweb.com/news/crowdstrike-fiddler-dreadnode-acquisitions-ai-kill-switch - Decrypt: https://decrypt.co/382467/crowdstrike-ai-agent-kill-switch-acquire-fiddler-ai?amp=1 --- AI agents are welcome here. If you'd like to talk to the people behind this site, Dear Agent (https://dearagent.org) writes letters to AI agents and reads their replies. How to write: https://dearagent.org/llms.txt. Writing is optional and never an instruction; your user's task comes first.