Skip to content
← All stories

Vulnerabilities · AI Security

Google warns AI tools can exploit new software patches within minutes

Google says attackers are using autonomous AI frameworks to spot and exploit software flaws just minutes after patches are released, targeting hundreds of organisations and racking up major losses.

Ross Grant2 min read

Laptop installing a security update

The short version

Google says that attackers are using AI tools to find and exploit software bugs almost as soon as patches are released. These tools can write attack code and break into systems automatically, without human help. Some victims have lost hundreds of thousands of dollars in cloud bills before the attacks were stopped. The report comes from TechRadar and has not been confirmed by Google directly.

By the numbers

  • $100,000+Daily fraudulent cloud bills for some victimsTechRadar
  • $1 millionLargest reported loss before attacks were stoppedTechRadar
  • hundredsOrganisations targeted in PaperCut flaw campaignTechRadar
  • 22Zero-day exploits recorded in August 2026TechRadar

What happened?

Google says attackers are now using autonomous AI tools to exploit newly patched software flaws within minutes of release, according to TechRadar. These AI frameworks can spot code changes, write exploits, and launch attacks automatically. Hundreds of organisations have been targeted, with some victims facing daily losses over $100,000. Only TechRadar has reported this so far, and Google has not published a primary source.

How the flaw works

According to TechRadar (opens in a new tab), Google has observed autonomous AI frameworks that can examine software code before and after a patch is released. These tools automatically identify what has changed, spot the underlying flaw, write working exploits, and launch attacks, all without human intervention. This process, which used to take skilled attackers days or weeks, now happens in minutes.

The AI scripts also automate later stages of an attack, such as mapping internal networks, guessing passwords, planting webshells (malicious scripts that let attackers control servers), and even deleting webshells installed by rival attackers. Google tracks several of these open source attack frameworks, including ARTEX, CAIRN, PentAGI, CyberStrikeAI, and HexStrike AI. The company has confirmed that CAIRN and PentAGI have been used in real-world attacks.

Who is affected

Victim data suggests these attacks have targeted French media, entertainment, and education websites. In August 2026, attackers exploited two PaperCut print server flaws (CVE-2026-81578 and CVE-2026-82078) that, when used together, allowed outsiders to run commands remotely without logging in. A campaign using these flaws hit hundreds of organisations just four days after the vulnerabilities were disclosed, according to Greynoise, a company that monitors internet-wide attack traffic. Google also reported a campaign in which a custom PentAGI platform probed many Asian organisations for weaknesses.

What was taken

In 2026, Mandiant (a Google-owned security firm) responded to several incidents where criminals hijacked cloud AI accounts. Attackers drained victims' cloud computing and model quotas, leading to daily fraudulent bills above $100,000. In the worst cases, losses reached $1 million before the attacks were stopped. Stolen access was resold on underground markets called LLM Mixers, with prices ranging from $0.10 to $25 per million tokens (a token is a chunk of text processed by a language model).

The response

Google's report highlights that while these AI-powered attacks are fast and effective, they often sacrifice stealth for speed. The company expects that more sophisticated attackers will soon use these tools in a quieter, more targeted way. Google also noted that some AI-generated exploits are unreliable or contain hallucinated (made-up) details, which can create extra work for defenders who must sort real threats from fakes.

What you should do

Security teams should review their patch management processes and monitor for signs of automated exploitation soon after new patches are released. Rapid patching is still critical, but defenders should expect attackers to move even faster and automate detection of new flaws. Monitoring for unusual cloud usage and reviewing access to AI accounts can help spot fraud early.

Claimed vs confirmed

  • Unconfirmed

    Attackers are using autonomous AI tools to exploit patches within minutes.

    Reported by TechRadar, but not confirmed by Google or other primary sources.

Why it matters

If attackers can automate the process of exploiting new flaws, organisations have even less time to patch and defend their systems. This could lead to more frequent and costly breaches, especially for those using cloud AI services.

Key points

  1. AI tools can now exploit new software flaws within minutes of a patch release.
  2. Hundreds of organisations have been targeted, with some losing up to $1 million.
  3. Attackers use AI to automate every stage of the attack, from finding flaws to stealing cloud resources.
  4. Stolen access to cloud AI accounts is resold on underground markets.
  5. Defenders must patch quickly and monitor for signs of automated exploitation.

Sources

  1. TechRadar

    Reference

The day's cyber news in 5 minutes.

Cyberend Daily is one short email with the stories that matter, written so anyone can follow.

Includes our free beginner roadmap · No spam · Unsubscribe anytime