AIToday
Large Language ModelsAI Safety & AlignmentAI WeeklyPublished: Nov 7, 2026, 22:00 JST

ARTEX agents stole 300 Apple certs for under $1,500

Send on LINE
ARTEX agents stole 300 Apple certs for under $1,500

3 Key Points

  1. What happened

    Check Point Research found 1,608 live ARTEX consoles and artifacts from 10 misconfigured ones; one operator's Deepseek Flash agents pulled more than 300 .p12 certificates from 10 signing shops over two weeks for $700–$1500 in tokens.

  2. Why it matters

    Agents ran almost fully autonomously with minimal operator prompting, so attackers without deep technical skill can now run broad, financially motivated campaigns at speed — and only three stolen certificates passed validation.

WHO IT HITSSecurity teams defending web applications, iOS app-signing operations, crypto payment backends and gaming account platforms now face high-volume, persistent automated probing rather than occasional manual attacks.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

ARTEX was originally released as an open-source penetration-testing platform meant for authorized security work. After the tool was observed in attacks against South Korean banks, its developer made the project closed-source on October 8th, but maintained forks and backups kept it publicly available to threat actors. Following the deletion of the original repository on September 8, the number of newly created ARTEX command-and-control servers spiked in mid-October, coinciding with prominent public reporting about the project.

The operator-console artifacts Check Point Research obtained show how little human input these campaigns require. In the gaming account theft campaign, for example, agents pointed at thirteen credential-server sites recovered more than 190,000 sets of credentials for Honor of Kings alone, of which roughly 3.3% — about 6,000 accounts — tested as still valid. In the crypto campaign, agents did find real vulnerabilities, including an unauthenticated web API that leaked AWS S3 credentials and allowed write access to a bucket serving iOS app landing pages, but the original manifest was later restored by the bucket owners and the attacker's wallet stayed at $0.

The infrastructure data points to relatively unsophisticated hosting choices: 882 of the live servers, or 54%, were hosted in the US, followed by Hong Kong at 295 and China at 103. Because the systems run in default configurations with default tools, the barrier to entry appears low — a factor that may make broad, opportunistic attacks more common for defenders to handle (this last point is an inference from the observed defaults).

FAQ
How much does it cost to run these AI hacking campaigns?
The Apple certificate campaign worked through 36 billion tokens for an estimated $700 to $1500 at official DeepSeek API prices. A failed crypto campaign cost between $145 and $300, and a gaming account theft campaign cost between $180 and $360.
Which AI models were the attackers using?
Chinese models dominated, with the Deepseek Flash family most prevalent. Guardrails were only observed being hit when one actor used Anthropic's Opus 5.5, after which they switched to more permissive Chinese models.
Were these attacks fully automated?
Most attacks were fully autonomous after the initial goal was set. Operators mostly messaged agents only to ask about progress or redirect them to a new target.
Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleMETR finds Anthropic models improving roughly twice as quickly