Google Discovers Autonomous AI Hacking Tool With Russian Code
Google spotted an autonomous hacking tool that scans for patches and writes exploits on its own. Crypto companies and development environments also appear to be especially vulnerable.

Key Takeaways
- Google discovered exploit_agent, a hacking tool that independently finds software bugs, writes exploits, and runs them without human help.
- The tool ran on an open staging server and included scripts for reconnaissance, brute-forcing, web shells, and collecting victim data.
- Google also saw North Korean actors using AI during intrusions at crypto companies in North America and Europe.
Google has discovered a hacking tool that independently finds and exploits software bugs, without human help. The find shows how quickly autonomous AI agents are starting to play a role in cyberattacks, including at crypto companies and other online services.
How exploit_agent Worked
According to Google, the tool was on an open staging server, a system attackers use to store and prepare their tools. Google called the tool exploit_agent. The software connected to multiple large language models, including commercial models and local variants, and monitored popular web software for patches.
As soon as an update appeared, the tool compared the code before and after the change to find the weak spot. It then automatically wrote an exploit and ran it. Google says this was the first time it saw attackers fully automate this process, from finding a flaw to exploiting it.
More Than Just an Exploit
The same server also held scripts for internal reconnaissance after a breach. Those let the tool map the network, guess passwords through brute-forcing, and place web shells so an attacker could regain access later. Google also found victim data there, with signs that the operator targeted French websites in media, entertainment, and education.
The server also contained Russian comments. Google translated them as instructions for automatic exploitation with LLM API keys, showing how directly AI was used in the attack chain. In practice, that makes cyberattacks faster and harder to stop, because vulnerabilities can be exploited soon after a patch is released.
Relevance for Crypto Companies
Google also says that in its research it saw North Korean actors using AI during intrusions at crypto companies in North America and Europe. That matters for the crypto sector because attackers are not only going after wallets or exchanges, but also development environments, code repositories, and access tokens. U.S. regulators and researchers had already warned earlier that autonomous AI tools can sharply speed up attacks.