
Most of the modern economy is renting space in a handful of giant clouds, and the biggest of them is Amazon Web Services. So it should get your attention that security researchers have tied a freely available AI hacking tool to real break-ins that walked out of AWS carrying live production secrets. The tool is called AntiFox, and it is a preview of what happens when the skill of a competent intruder gets packaged into software anyone can download and run.
What AntiFox is
AntiFox is a fork of an offensive AI tool called ARTEX, with a Chinese-language interface. It is an agentic penetration-testing platform, meaning it chains together the steps a human hacker would normally do by hand: mapping exposed services, probing endpoints, testing credentials, and writing up what it found. Penetration-testing tools exist so that defenders can attack their own systems and fix the holes. This one is being pointed the other way, at other people’s systems, and it does the work at machine speed.
The break-in
Arctic Wolf Labs linked infrastructure running AntiFox to a set of intrusions. The attackers used IIS web shells and valid AWS credentials to enumerate cloud services and reach into production secrets. In one documented case most of the requests came back denied, but the sequence included a successful call that pulled a live secret straight out of AWS Secrets Manager. The infrastructure was reportedly run on Alibaba Cloud. Arctic Wolf has been careful to say the link supports attribution to AntiFox-related activity rather than proving an AI agent independently performed every step, which is the honest caveat the scarier headlines leave out.
There is no AWS flaw here, and that is the point
This matters: there is no reported vulnerability in Amazon’s software. The attackers used valid credentials, the kind that leak out of config files, public repositories and sloppy deployments every single day. So this is not a story about a cloud giant’s code failing. It is the oldest problem in security, a leaked key, meeting a new multiplier. In a lab test, researchers at GitGuardian ran an ARTEX build with no secrets scanner against nine planted credentials, and it found eight of them in six to thirteen seconds. A human rummaging by hand takes a great deal longer. The AI does not find new doors. It finds the unlocked ones faster than anyone can close them.
Why this is the sequel we were promised
ARTEX, the tool AntiFox was forked from, went closed-source on 8 October after being spotted in attacks on South Korean banks, and Microsoft separately reported attackers using it to hunt exposed GitHub credentials. Shutting the source did nothing useful, because AntiFox is the fork that got out first. That is the whole pattern in one example: offensive AI tools leak, fork, spread and get cheaper, and each generation lowers the bar for the next set of attackers. The skill that used to live in a few expensive heads now lives in a download.
What a business should actually do
The defences are unglamorous and they work. Stop storing long-lived AWS keys in config files and code repositories. Use short-lived credentials and roles instead. Turn on secrets scanning in your repositories and build pipelines. Rotate anything that has been exposed. Lock down permissions so that a single stolen key cannot reach production secrets in the first place. And watch for unusual Secrets Manager activity, because that odd 3am call is the tell. None of this needs its own AI to pull off. It needs doing.
What this means
The headline is not that AWS broke, because it did not. The headline is that the only thing standing between an attacker and the cloud your business lives inside has always been a few credentials and the time it takes to find them, and AI has crushed that time down to seconds. AntiFox is one tool. There will be more, cheaper and sharper, because that is the only direction this goes. The cloud runs the economy, and the lock on the door is still, far too often, a password somebody left in a text file.
Edgar Friendly, calling the AI bullsh*t. More tools, more swearing: toptoolstack.com.
Did you know? A huge share of cloud breaches start not with a clever exploit but with a credential that was sitting in plain sight. The glamorous zero-day gets the movie. The leaked key does the actual damage.
Sources: BleepingComputer; The Hacker News; GitGuardian; Security Boulevard.