AI Trends

Google Warns That Fully Automated AI Cyberattacks Have Entered a New Phase

cnBetaUpdated Nov 17 14:11

AI summary

Google's Threat Intelligence Group (GTIG) reported that it has for the first time observed threat actors using large language model frameworks to complete vulnerability discovery and exploitation with no human involvement at all, and predicted that within the next three months hackers will be able to launch low-barrier attacks requiring almost no human effort. The report named open-source penetration testing platforms such as CyberStrikeAI, PentAGI, and CAIRN as being misused, and also mentioned a multi-agent system called exploit_agent that carried out automated attacks on French media, entertainment, and education websites. Google recommends that enterprises promptly patch internet-facing systems and shorten patch deployment intervals, and introduced the Big Sleep security agent and the AI Threat Defense architecture used for defense.

Why read this

AI score 78

Google's threat intelligence team has laid out a timeline for AI agents to automatically find and exploit vulnerabilities, and named attack frameworks that have already emerged, which security teams can use as a reference when assessing patch windows.

Topics

Tags

Body · AI translation

The report shows that attackers can now use AI agent systems to automatically carry out vulnerability discovery, analysis, and exploitation, greatly shortening attack preparation time and improving operational efficiency. GTIG researchers predict that within the next three months, hackers will be capable of launching low-barrier cyberattacks that require almost no human involvement, with most of the process completed autonomously by AI agents.

Researchers found that cybercriminals are heavily exploiting legitimate open-source penetration testing tools and frameworks to carry out attacks, including platforms such as CyberStrikeAI, PentAGI, and CAIRN. Not long before this, the U.S. Federal Bureau of Investigation (FBI) also issued a warning that an AI-driven penetration testing platform called ARTEX is being used by threat groups to automate attacks against enterprises and institutions worldwide.

GTIG said that in the most recent quarter, its research team observed for the first time threat actors using large language model frameworks to complete the entire process of vulnerability discovery and exploitation with no human involvement at all. This means automated cyberattacks have crossed an important threshold and entered a new stage of development.

The report cited an example in which researchers found hackers using a multi-agent AI system called "exploit_agent" to carry out automated attacks on French media, entertainment, and education websites. The system can continuously analyze software updates for mainstream web applications, look for potential security vulnerabilities in them, and automatically attempt to launch exploits.

Google noted that the emergence of such frameworks has greatly compressed the time window for security teams to patch vulnerabilities. In the past, security personnel often had relatively ample time to deploy updates and fix vulnerabilities; with the help of AI automation tools, attackers can weaponize known vulnerabilities faster and thus launch attacks quickly. However, Google also said that these systems sometimes produce misjudgments or invalid attack leads.

GTIG believes that attackers' adoption of such frameworks marks a major shift in the cyber intrusion model. Although the basic principles of intrusion have not changed—for example, traditional methods such as stealing credentials or exploiting unpatched systems still exist—automated frameworks can chain together complex attack chains at "machine speed" and conduct operations on a larger scale, which brings unprecedented new challenges to cyber defense efforts.

In response to this trend, Google emphasized that enterprises should patch internet-facing systems as soon as possible and shorten the time between patch release and formal deployment as much as possible. The report argues that, compared with actively searching for brand-new vulnerabilities, attackers now prefer to quickly exploit vulnerabilities that have already been publicly disclosed but not yet patched, so keeping systems continuously updated remains one of the most effective measures for improving security.

At the same time, Google believes that AI is not only being used for offense but should also be used for defense. The company mentioned an AI security agent called "Big Sleep," which can proactively find known software vulnerabilities and assist in fixing them. Google also introduced its AI Threat Defense security architecture, which combines AI models, risk prioritization analysis, and automated remediation capabilities to detect and respond to cyber threats faster.

Industry insiders believe that as AI-driven attack tools continue to mature, the cybersecurity field is entering a new stage in which both attackers and defenders rely on artificial intelligence. In the future, enterprises will not only need to face the pressure brought by the expansion of automated attacks, but must also use more advanced AI defense systems to cope with continuously escalating cyber risks.

Source

cnBeta