Skip to content
Technology Editorial

Google Warns of Fully Automated AI-Powered Cyberattack

Google identified a fully automated cyberattack using AI to find software vulnerabilities, develop exploits and execute attacks without human involvement.

Google Warns of Fully Automated AI-Powered Cyberattack
Ishant Sharma
Technology / Science Contributor
Published: 5 min read
Reported and edited under BigBreakingWire editorial standards.
Advertisement

Google’s Threat Intelligence Group (GTIG) has identified what it describes as the first documented case of a fully automated cyberattack exploiting software vulnerabilities without human involvement during the attack process. The discovery highlights how attackers could use artificial intelligence to identify weaknesses in software, develop exploits and execute attacks at greater speed.

Researchers found an exposed staging server running a multi-agent framework called exploit_agent. The system connected multiple advanced and locally hosted AI models to automate different stages of vulnerability exploitation.

Advertisement

The findings come as Google reports a rise in vulnerability disclosures and exploits during 2026. The company warned that AI could shorten the time between the release of a security patch and the development of an exploit targeting the vulnerability.

AI Framework Automates Vulnerability Exploitation

According to GTIG, the exposed server contained a framework designed to compare software code before and after security patches. It could use those differences to identify potential vulnerabilities, develop exploits and execute them automatically.

The server also contained stolen victim data and scripts designed to carry out additional malicious activity. These included brute-forcing systems, installing webshells and removing tools deployed by rival attackers.

Google’s telemetry suggested that the operator targeted websites in the French media, entertainment and education sectors.

The company said vulnerability disclosures and exploits nearly doubled between January and August 2026. Zero-day exploits increased to 22 in August, adding to concerns about how quickly attackers may be able to exploit newly identified weaknesses.

GTIG and Mandiant also identified threat actors misusing open-source AI-powered penetration-testing frameworks. The tools identified included ARTEX, CAIRN, CyberStrikeAI and PentAGI.

Reported targets included US aerospace organisations and businesses across Asia. Google also identified a group with links to China that was developing its own automated attack framework, potentially for espionage operations.

Cybersecurity Researchers Warn of Expanding AI Misuse

Google said the automated tools currently prioritise speed over stealth. Their activity can generate network traffic and technical traces that may help security teams identify malicious operations.

However, the company warned that these capabilities could become harder to detect as the technology improves. Automated systems can reduce the amount of direct human involvement needed to identify vulnerabilities and carry out attacks.

Google said it had disabled accounts and assets linked to the misuse of its Gemini models. It also strengthened its security classifiers based on the findings.

Advertisement

The concerns extend beyond vulnerability exploitation. A separate reported incident involved malicious JavaScript delivered to visitors of websites belonging to major organisations, including Zoom, HSBC, Equifax, CNN, Salesforce and IKEA.

Security researcher Scott Helme said the scripts reached visitors across 178 organisations in multiple sectors through Optimizely, a website optimisation platform. The reported activity included attempts to replace cryptocurrency payment addresses with attackers’ wallet addresses, capture credit card details, steal browser cookies and screenshots, and deploy exploit code on iOS devices.

Cybersecurity firm HudsonRock also reported finding evidence linked to the compromised websites. Salesforce said it was investigating unauthorised changes to Optimizely scripts and had disabled them as a precaution.

Optimizely, Apple, Zoom, HSBC, Equifax, CNN and IKEA did not immediately respond to Reuters’ requests for comment, according to the supplied material.

The two incidents demonstrate different risks involving malicious software activity. Google’s findings concern the automation of vulnerability exploitation, while the website incident involved scripts that could target visitors through compromised third-party website code.

Source attribution: Google’s Threat Intelligence Group and Mandiant findings; statements attributed to security researcher Scott Helme, HudsonRock and Salesforce; Reuters reporting as provided in the source material.

BBW Analysis

What It Means

AI can reduce the time and human effort needed to identify software weaknesses and turn them into working exploits. This creates pressure on organisations to identify vulnerabilities and secure systems quickly. The separate website incident also shows how third-party scripts can expose visitors to risks involving payment details, cryptocurrency transactions and personal browsing information. Google said the automated tools currently leave technical traces that can help security teams detect them, but warned that this could change as the technology develops.

What to Watch

  • Further findings from Google’s Threat Intelligence Group.
  • Whether AI-powered exploitation becomes harder to detect.
  • Additional details about the exploit_agent framework.
  • Security investigations into the compromised Optimizely scripts.
  • Further action against accounts and assets linked to AI misuse.

Market & Investor Impact

  • Cybersecurity Risk: Automated exploitation — AI tools could reduce the time required to identify and exploit software vulnerabilities.
  • Website Security: 178 organisations — Malicious scripts reportedly reached visitors across organisations in multiple sectors.
  • Payment Security: Cryptocurrency and card data — The scripts attempted to redirect cryptocurrency payments and capture card details.
  • Enterprise Response: Salesforce disabled scripts — The company took precautionary action while investigating unauthorised changes.
  • Threat Detection: Technical traces remain — Google said current automated tools generate activity that may help security teams detect attacks.

What did Google discover about automated cyberattacks?

Google’s Threat Intelligence Group identified a framework called exploit_agent that could compare software code, identify potential vulnerabilities, develop exploits and execute them automatically without human involvement during the attack process.

How does the exploit_agent framework work?

The framework connects multiple advanced and locally hosted AI models. It reportedly examines code before and after security patches to identify potential weaknesses, develop exploits and execute them.

Which organisations were affected by the separate malicious script incident?

Websites belonging to Zoom, HSBC, Equifax, CNN, Salesforce and IKEA were among those reportedly affected. Security researcher Scott Helme said malicious scripts reached visitors across 178 organisations through Optimizely.

What information did the malicious scripts attempt to steal?

The scripts reportedly attempted to capture credit card details, steal browser cookies and screenshots, replace cryptocurrency payment addresses and deploy exploit code on iOS devices.

What action did Google take after its findings?

Google said it disabled accounts and assets linked to misuse of its Gemini models and strengthened its security classifiers. The company also warned that automated attacks could become harder to detect as the technology improves.

Advertisement
Sponsored
Ishant Sharma
Technology / Science Contributor

Ishant Sharma

Ishant Sharma is a Technology and Science Contributor at BigBreakingWire, covering artificial intelligence, software, semiconductors, gadgets, and emerging technologies.

Last reviewed by Ishant Sharma on November 17, 2026

Disclaimer: BigBreakingWire provides news and informational content for general educational purposes only. The information presented on this website does not constitute financial, investment, tax, or legal advice. Readers should consult qualified professionals before making any financial decisions. BigBreakingWire, its authors, and editors are not responsible for any financial losses or damages arising from the use of information on this site.

Discussion

Share your take — react and comment in seconds, no login needed.

0 comments

Loading discussion…

Be civil — abusive or promotional comments get removed. No login required.

Most Read

Install app