AI-powered cyberattacks grow more sophisticated as tech giants launch defenses

The rise of autonomous hacking

Cyberattacks are becoming more sophisticated as artificial intelligence (AI) assists in discovering vulnerabilities and developing attack code. These tools allow complex intrusions to be carried out at a lower cost than previous methods. Recent incidents show AI agents analyzing situations to select the specific commands needed for a breach.

In one instance, a security company reported that hackers used an open-source AI tool called CyberStrikeAI to target educational institutions. The AI agent was able to establish persistent access and complete initial reconnaissance within two minutes of deployment. It then spent several days compromising domain administrator accounts to steal credentials and exfiltrate databases.

This setup allows human attackers to specify high-level goals while the AI handles the technical execution. The AI selects the appropriate tools for the specific environment, and a program on the victim's system runs the commands. This automation significantly speeds up the final stages of an attack, which in some cases were completed in less than an hour.

AI in real-world exploits

Hackers have also used AI to develop intrusion tools for specific hardware targets. For example, suspected China-linked hackers used an attack platform called AIScan to target firewalls. This platform integrated Anthropic's Claude Code with other Chinese AI models to identify software versions and write tailored attack code.

Another notable case involved the cryptocurrency exchange Binance, where attackers exploited internal AI tools. The hackers first tricked an engineer into installing a malware-infected AI coding tool. They then used that foothold to access the exchange's in-house AI systems and gain control over employees responsible for wallet operations.

By abusing these internal tools, the attackers were able to discover tokens that granted them privileges to deploy malicious programs. They eventually used these credentials to execute scripts that transferred cryptocurrency to external wallets. These attacks demonstrate how even internal corporate AI tools can become pathways for malicious actors.

The corporate defense response

Major AI developers are now positioning cybersecurity as a primary growth engine for their enterprise solutions. OpenAI recently hired a security executive to bolster its sales to corporate customers. The company's CFO stated that cybersecurity would be a major driver of the company's future growth.

OpenAI launched a system called Daybreak to help companies detect and fix security flaws. Daybreak analyzes software code to identify potential vulnerabilities and verifies if they can actually be exploited. It also supports the generation and testing of code fixes to help developers secure their systems more quickly.

In addition to Daybreak, OpenAI introduced a security-focused model called GPT-5.6-Cyber. This model is part of a broader effort to provide tools that can automatically find and patch bugs. The goal is to create a proactive defense that can keep up with the speed of AI-driven attacks.

Collaborative security programs

Anthropic and Google are also expanding their presence in the security market with specialized tools. Anthropic released Claude Code Security, which analyzes a company's entire source code to find vulnerabilities that traditional tools might miss. It assesses the severity of these flaws and proposes specific fixes for developers.

Anthropic also launched Project Glasswing in partnership with companies like Amazon Web Services and Microsoft. This joint program helps major technology and financial firms use AI to proactively identify and fix security issues. Participating organizations reported finding more than 10,000 high-severity or critical vulnerabilities in just over a month.

Google launched a similar initiative called Fairwind for government agencies and major companies. This program provides limited access to Gemini-family models to help organizations automatically search for software vulnerabilities. These models have already been used by security firms to discover critical flaws in web servers and validate various fixes.

What comes next for security

The security landscape is shifting toward an automated arms race between offensive and defensive AI. As hackers use AI to find