中国建造的AI黑客机器让任何白痴3小时内渗透系统,FBI警告
Chinese-Built AI Hacking Machine Lets Any Idiot Infiltrate Systems In Under 3 Hours, FBI Warns

原始链接: https://www.zerohedge.com/technology/chinese-built-ai-hacking-machine-lets-any-idiot-infiltrate-systems-under-3-hours-fbi

ARTEX 是中国开发的 AI 黑客框架,FBI 警告称它让低技能攻击者能够以机器速度自动化复杂入侵。自 8 月以来,攻击者已在全球多个行业的组织中使用它。它利用外部 AI 模型(通常是中国模型,如 DeepSeek)将自然语言目标转换为命令。尽管其 GitHub 仓库已于 10 月移除,ARTEX 仍通过分叉、变体和商业版继续传播。事件包括 10 月的 ConnectWise Cloud Backup 入侵,以及对韩国金融机构和银行的攻击。研究人员发现超过 1600 个暴露的控制台,谷歌表示此类工具以隐蔽性换取速度。官员警告 AI 驱动的网络威胁正在升级,并敦促组织减少暴露的系统,特别是被忽视的合作伙伴、遗留和第三方服务。

相关文章

原文

ARTEX, an AI hacking framework developed in China, lets attackers "with minimal technical expertise" run complex, multi-target break-ins at machine speed, the FBI warned in a FLASH advisory last week. In some cases, attackers went from initial targeting to privileged access in under three hours, while pulling ARTEX off GitHub didn't stop it from spreading.

Since at least August, the bureau says, a range of cyber actors have used ARTEX against organizations worldwide in finance, government, technology, defense, semiconductors, energy, healthcare, manufacturing, telecommunications and law. "As of early November 2026, researchers identified over 1,600 publicly exposed ARTEX console instances," the advisory says, "with total deployments likely significantly higher when including systems behind firewalls or in private networks."

ARTEX is a multi-agent platform written in Go that turns a goal typed in plain language into commands, which its agents then carry out with little further input from the operator. It has no AI model of its own: it plugs into outside ones, from Anthropic's Claude and OpenAI's GPT to DeepSeek, GLM, Grok and Kimi. On the attackers' own machines, though, Chinese models are the most common, with DeepSeek's Flash models dominating thanks to their capability, low cost and "lack of observable guardrails when framed for penetration testing," according to the FBI.

Treasury Secretary Scott Bessent went further on Friday, telling CNBC that AI attacks are all coming from Chinese models, and that those models are dangerous.

At a Heritage Foundation event on Nov. 12, Bessent said Iran had used "a Chinese model" in the cyberattacks that knocked out refrigeration at French supermarkets, said Chinese models are dangerous because "they don't have guardrails on them," and warned that "you're probably going to see a big incident somewhere in the world." He offered no evidence for the Chinese-model link, and Chinese state media denied it.

Bessent's case against Chinese models goes back further. In July he threatened sanctions after saying the administration was finding watermarks of US large language models on many Chinese ones, and Jensen Huang's ouster last week took out open AI's most powerful defender just as Washington eyes a crackdown on open-weight models.

The Wall Street Journal identified ARTEX's creator as Li Puhua, a Chinese cybersecurity engineer who goes by "Autumn." Chinese cybersecurity firm ThreatBook says ARTEX began as a personal project by one of its engineers. According to the FBI, Li released ARTEX on GitHub in July under the handle "Autumn-27" for authorized security testing, then took the repository down on Oct. 8 after reports of abuse. By then the project had passed 12,000 forks, Datadog's security researchers found. The FBI says ARTEX lives on through forks, archived copies, local variants and a commercial edition, ARTEX PRO.

The best-known case so far is the October compromise of ConnectWise Cloud Backup. Microsoft attributes it to Storm-0940, which it calls "a China-state aligned actor," and says the group used ARTEX "to automate attacks targeting hundreds of organizations globally" and tried to sign in to more than 2,250 downstream tenants. According to Microsoft, the attack did not come through a compromise of, or a flaw in, its own systems.

Beyond Beijing-aligned spies, the FBI cites attacks on South Korean financial institutions that stole personal information on about 67,000 people, and Microsoft has tracked a Cambodia-based, financially motivated group, Storm-3358, using ARTEX against banks and other financial firms since October. In AWS cloud accounts it watched that month, Datadog saw ARTEX-driven attacks hunting for financial data, and it found hundreds of exposed ARTEX consoles, many with Mandarin interfaces and hosted in places like Singapore and Hong Kong.

Google's Threat Intelligence Group, which counts ARTEX among a new wave of AI attack frameworks along with CAIRN, Hexstrike and PentAGI, calls it "a paradigm shift towards intrusions occurring at machine speed." There is one catch for the attackers, in Google's words: "they currently trade stealth for velocity."

"Imagine giving a Terminator a mission in plain English," Christopher Walcutt, chief information security officer at DirectDefense, told the New York Post, adding that an attacker can now run it from a phone. Scott Schober, CEO of Berkeley Varitronics Systems, told the paper that cyber criminals are sometimes ahead of the curve, and that attackers have "just got to get in once."

The FBI's Dallas office posted about the advisory on Tuesday, in the middle of a week of outages, and the post spread as NSA, Apple and Equinor systems were hit by disruptions. Hours later the FBI's main account said the advisory "is not a new announcement, and the FBI has not issued any additional warning since," and Director of National Intelligence Jay Clayton has said the outages were not a cyberattack, as we reported Thursday, when news broke that White House National Cyber Director Sean Cairncross is leaving.

Bessent's "big incident" warning came the same day Microsoft went down. The run so far:

  • Nov. 12: Microsoft's outage spreads worldwide; the company blames "an issue during an earlier network capacity upgrade"
  • Nov. 13: The FAA orders a nationwide ground stop after its communications system crashes; its administrator has since taken the blame
  • Nov. 16: JPMorgan customers are locked out by a "technical issue" the bank says shows "no evidence" of being cyber-related
  • Nov. 17: The NSA's classified internal network goes down for "routine maintenance," Apple reports service disruptions, and Norway's Equinor confirms an ongoing cyberattack
  • Nov. 18: PlayStation reports disruptions during the GTA VI rollout, and the ransomware gang The Gentlemen reportedly hacks Nvidia
  • Nov. 19: Hackers expose some of JPMorgan's richest customers

Two senators, Democrat Maria Cantwell and Republican Mike Rounds, warned the Trump administration on Wednesday that AI-enabled cyber threats are "rapidly escalating" and outpacing America's defenses.

The FBI's advice starts with shrinking what attackers can see from the outside, because ARTEX "excels at" finding the systems nobody watches closely: partner and supplier portals, mobile backends, legacy services and third-party integrations.

联系我们 contact @ memedata.com