You don't need to know how to hack anymore if you can point a free AI agent at a target and wait. The FBI and Google both say criminals and state-backed crews are already doing it, and your credentials are what they find first.
The FBI says a single open-source framework called Artex has taken attackers from first targeting to privileged network access in under three hours. This week Google's Threat Intelligence Group named it alongside Cairn, HexStrike AI, CyberStrikeAI and PentAGI, according to CIO Dive, and described attackers moving from typing prompts to running coordinated, multi-agent frameworks that need little human supervision.
Axios laid out the mechanics on November 18. A hacker gives the framework a target and an objective, such as getting into a company's network. AI agents draft the attack plan, hunt for weaknesses, run the hacking tools to test entry points, share what they find and change tactics when something fails. They don't need a human watching, and they don't sleep.
Artex began life as a penetration-testing platform for security teams. It's written in Go and plugs into outside large language models, according to IBTimes Australia, which reported that a developer going by Autumn-27 posted it on GitHub in July and took it down on October 8 after abuse reports. Forks, mirrors and a paid version called Artex PRO are still around.
The FBI's FLASH alert, issued November 13, says that since at least August a range of malicious actors have used Artex against banking, government, technology, defense, semiconductor, energy, healthcare, manufacturing, telecom and legal targets. The same coverage notes South Korean financial institutions lost personal data on 67,000 people, and that Microsoft tied a campaign against more than 2,250 ConnectWise customers' Microsoft 365 environments to a group it calls Storm-0940, which it links to the Chinese state. More than 1,600 Artex control consoles were sitting open on the internet in early November.
Researchers find 1,465 exposed Artex AI hacking consoles open on the web Censys found 1,465 Artex consoles exposed online as of November 4, with 32 open directories leaking Windows lateral-movement tools, a sudo exploit and Linux implants that anyone could grab and reuse. - exposed Artex AI hacking consoles found on internet - Chinese AI penetration testing tool security breach discovery
Check Point Research, reported by Tech.co on November 9, counted over 400,000 messages across 292 targets in one Artex campaign after the attacker gave only the initial setup. Check Point said the agents scan, exploit and even reason through supply chain attacks on their own.
What it costs an attacker #
Very little. Research from Vals, cited in the same Tech.co piece, put the token bill for an attack run on DeepSeek at $2 to $3, against $17 to $46 on OpenAI's GPT-6-Sol, with task completion of 73% versus 81%. The Next Web adds that attackers favor DeepSeek's Flash models for exactly that price and their thin safeguards.
Frankly, that's the number to hold onto. A skilled intruder used to be the expensive part of an attack. Now the expensive part is a few dollars of compute, and anyone can pay it.
It doesn't always work. Check Point described one crypto-focused campaign that burned through 167.5 million tokens, found several vulnerabilities and still earned nothing. Its summary was that attacks mostly look the same with AI, they're just faster and bigger.
What to fix first if you run a small company #
Faster and bigger means the gaps that used to sit unnoticed for months get found in an afternoon. Look at how the Artex cases actually went. At Barracuda, the agents found an exposed third-party credential that hadn't been revoked. It reset every customer credential as a precaution. Its CISO told The Next Web that defenders who rely on spotting malicious activity and then responding will often arrive after the attacker has finished.
Amazon's decoy sensors, which tracked Artex agents between October 3 and November 5, caught one campaign stealing API keys for AI services and using them to pay for its own operations. So start with the boring list. Rotate and revoke old keys and vendor tokens, find secrets sitting in code repositories, lock down admin consoles, and check who holds privileged access to your Microsoft 365 tenant. If a ConnectWise-style supplier can reach your cloud, assume the agents will test that door too.
There's a market in this for defensive AI vendors, and it will get crowded fast. The same tools cut both ways: Artex and Cairn started as security-testing software, and nothing stops you from running one against your own domain before someone else does. Do that before you sign a contract for a product that promises to detect the attack. Barracuda's experience suggests detection alone arrives late.
Oracle and OpenJDK patch Java4Shell, a critical Java flaw found with AI Calif, a US security startup, says Java4Shell can let attackers run code on servers using OpenJDK 17, 21, 25 or 27 on AVX-512 chips, including Jenkins, Jira and Keycloak. ARM64 machines aren't affected, and Oracle's severity score is lower than Calif's. - how to patch Java4Shell CVE-2026-114938 - Java Base64 vulnerability on Intel AMD servers
The FBI's alert, Google's report and Amazon's sensor data all point the same way: faster, cheaper attacks run by agents that need little supervision.
Also read: Oracle and OpenJDK patch Java4Shell, a critical Java flaw found with AI • Take-Two shares jump 7.5% as GTA VI hits 99 on Metacritic a day before launch • Palmer Luckey reveals SLAB, a seastead he says he has funded since selling Oculus
This article is posted in Technology News, check it out for more related stories.
Join the discussion #
Open in the community → Almost there. Sign in and your reply posts straight away.