Google Threat Intelligence Group has published an LLM Maturity timeline, highlighting the evolution of the use of AI in cyberattacks.
GTIG highlights that attackers are using AI to reverse engineer security patches
Photo Credit: Unsplash/ Igor Omilaev
Google Threat Intelligence Group (GTIG), the cybersecurity research arm of the Mountain View-based tech giant, has published a new report with Mandiant, highlighting the risks AI agents pose to cybersecurity. GTIG claims that autonomous cyberattacks executed by AI agents will rise sharply by next year. The group has observed a shift to autonomous workflows and multi-agent frameworks by various threat actors for carrying out attacks against the digital infrastructures of companies and other institutions. Hackers are also increasingly using AI tools to reverse engineer security patches to exploit vulnerabilities, GTIG claims in the new report.
In a new report, GTIG has published its observations on how bad actors are abusing large language models (LLMs) to execute cyberattacks. The group has observed that hackers are increasingly using generative AI tools to shorten cyberattack timelines, scale operations, and carry out multi-stage intrusions faster. GTIG has noticed multiple trends in the last quarter. The group highlights that there has been a shift to deploying multi-agent offensive frameworks, like ARTEX, CAIRN, CyberStrikeAI, and PentAGI, to execute intrusions at a “greater operational velocity”.
According to GTIG's evolution timeline for the use of LLMs by threat actors, the world is currently at Stage 5, where LLM-powered tools execute multiple attack lifecycles. The cybersecurity research group warns that in 1 to 3 months, the evolution cycle will reach Stage 6, where threat actors will start executing multi-agent-driven, unsophisticated cyberattacks with minimal human interaction. Lastly, Stage 7 is predicted to be 3 to 6 months away, where bad actors will be able to carry out sophisticated, fully autonomous cyberattacks by using AI agents.
GTIG says that attackers are abusing “open-source, agentic penetration-testing (pentesting) frameworks” to automate discrete phases and specific operational tasks for executing attacks. An orchestration agent is used for doing reconnaissance and scanning external networks to gain initial access. Later, the AI agent can establish persistence to gain greater control over the network. It can then begin carrying out internal reconnaissance and laterally breach other branches to compromise systems.
With an increase in operational scale, autonomously executed cyberattacks can have a lower success rate. However, in its assessment, GTIG observes that the broader reach and frequency of cyberattacks carried out by AI agents will compensate for the fall in success rate.
Further, GTIG says that bad actors are increasingly using LLMs and agentic frameworks to automate the reverse engineering of security patches with the help of differential analysis. The group warns that this allows the attackers to weaponise n-day vulnerabilities or security gaps that have already been observed, while also eliminating the need for human intervention in scaling exploitation pipelines against internet-facing servers.
The report also highlights that financially motivated attackers are using “LLMJacking” to profit. Here, threat actors steal LLM credentials and cloud resources from vulnerable enterprise environments to resell access to high-performance AI models via proxy services, like LLM Mixers. Similarly, state-sponsored threat actors have begun routinely using LLMs to generate “synthetic media” at scale and create localised personas, while also automating their distribution across platforms.
Get your daily dose of tech news, reviews, and insights, in under 80 characters on Gadgets 360 Turbo. Connect with fellow tech lovers on our Forum. Follow us on X, Facebook, WhatsApp, Threads and Google News for instant updates. Catch all the action on our YouTube channel.