Google says hackers now use AI agents to find and exploit software flaws on their own

Google says hackers now use AI agents to find and exploit software flaws on their own

Google's threat research team says it has, for the first time, seen attackers use AI agents to find and exploit software flaws without step-by-step human direction. The finding was published on Monday by the Google Threat Intelligence Group (GTIG), which tracks online threats to Google and its customers.

AI agents are software programs powered by large language models (LLMs), the technology behind chatbots, that can carry out tasks on their own. In this case, Google says attackers used them to automate parts of hacking that previously needed people doing the work.

What Google's researchers found

In the past quarter, GTIG identified threat actors using LLM frameworks to fully automate vulnerability discovery and exploitation without a human in the loop, according to the report. This is a change from Google's September report, which said it had not yet observed threat actors deploying fully autonomous pipelines against targets in the wild.

The researchers described a tool found on an exposed staging server. The tool compared software versions, identified potential flaws, and generated and executed exploit code. The server hosted a multi-agent framework set up to run continuous "differential analysis" — comparing code before and after patch announcements on popular web applications — so the model could autonomously discover and exploit vulnerabilities.

Why patching windows are shrinking

Google warned that these automated workflows give organizations less time to apply security patches before attackers find ways to exploit vulnerable systems. Attackers are using AI agent frameworks to automate patch reverse engineering through differential analysis and to weaponize "N-day" vulnerabilities, which are publicly known security flaws.

The report also warned that LLMs are helping attackers scale exploitation pipelines against internet-facing servers without human intervention.

Key numbers from the report

  • Vulnerability disclosures more than doubled between January and August, according to Google.
  • Exploited zero-day vulnerabilities — flaws with no patch available — averaged eight per month in 2025.
  • That average rose to 11 per month from January through August 2026, reaching 22 in August.

Fast-moving attacks and other warnings

The automated workflows also speed up how quickly attackers move from initial access to gaining domain administrator control, a high level of access to a network. The report cited an August 2026 campaign against PaperCut, in which attackers reached that level of control within minutes, according to cybersecurity firm GreyNoise.

Cloudflare separately reported attempts to exploit PHP flaws against its customers in October. GTIG said the two PHP vulnerabilities involved were discovered using LLM analysis, because technical information published before the Oct. 21 patch let researchers and attackers study the flaws and attempt exploits.

AI company Anthropic issued a similar warning in a September report, saying AI-assisted attacks have evolved into multi-agent operations that automate most of the cyberattack cycle, from discovery to exploitation and data theft. Anthropic described threat actors running "multi-victim campaigns using agentic AI that would previously have required teams of operators."

What this means for security teams

The practical effect, according to Google, is that organizations have less time to install security updates before attackers can break in. As AI tools automate both finding flaws and writing exploit code, the gap between a patch being released and an attack being possible is narrowing.

Sources

Newisty Editorial Team
Written by

Newisty Editorial Team

Technology · Crypto · Digital Economy
View all posts

Newisty Editorial Team covers technology, cryptocurrency, digital products, online platforms, developer tools and the wider digital economy. Our content is researched from official sources, company announcements, public documentation, market data and other primary or reputable sources. Articles are reviewed and edited before publication for clarity, accuracy and useful context.

Comments (0)

Leave a comment
Your comment will appear publicly after submission.
No comments yet. Be the first to comment!