Google Reveals AI‑Driven Hackers Automate Vulnerability Discovery and Exploitation
Google’s Threat Intelligence Group uncovered attackers using large language models to autonomously find and exploit software flaws, raising new security concerns for Asian firms.
Key takeaways
- Google identified a staging server hosting a multi‑agent framework that automatically discovers and exploits software vulnerabilities.
- The number of zero‑day exploits rose from an average of eight per month in 2025 to 11 per month in 2026, peaking at 22 in August.
- Attackers used the framework to gain domain administrator access within minutes, as seen in the August 2026 PaperCut campaign.
Google’s Threat Intelligence Group (GTIG) released a report on Monday detailing how attackers are now leveraging large language model (LLM) frameworks to automate the entire cycle of vulnerability discovery and exploitation without human intervention. The researchers identified a staging server that hosted a multi‑agent system designed to perform continuous differential analysis on popular web applications, comparing code before and after patch releases. According to GTIG, the model can autonomously identify potential flaws, generate exploit code, and execute it against vulnerable targets. This marks the first documented instance of fully autonomous LLM‑powered attack pipelines in the wild.
Automated Vulnerability Pipeline
The exposed staging server, as described by GTIG, was configured to conduct differential analysis on software versions, effectively reverse‑engineering patches to uncover new weaknesses. The AI agents could then weaponise these findings, creating and deploying exploit code without human oversight. Google warned that such automated workflows give attackers a significant time advantage, allowing them to develop and test exploits before organizations can patch vulnerable systems.
Impact on Zero‑Day Exploits
Google’s September report noted a doubling of vulnerability disclosures between January and August. In 2025, the average number of zero‑day exploits per month was eight, rising to 11 in 2026 and reaching 22 in August alone. The report cited the August 2026 PaperCut campaign, where attackers achieved domain administrator control within minutes of a vulnerability being discovered, illustrating the speed of the new automated attack model.
Industry Reactions
Cybersecurity firm GreyNoise highlighted the rapid progression from initial access to full control in the PaperCut incident. Anthropic, an AI company, also warned that AI‑assisted attacks have evolved into multi‑agent operations that can automate most of the cyber‑attack cycle, from discovery to data theft. The Defiant and other outlets have echoed these concerns, noting that the rise of AI‑driven exploitation could lead to a “vulnerability apocalypse” for internet‑facing servers.
What to Watch Next
Asian technology firms and regulators will need to reassess patch management and incident response strategies in light of these findings. Security teams should consider implementing AI‑based monitoring to detect anomalous exploitation attempts early. Industry bodies may push for stricter guidelines on AI usage in cybersecurity, and governments could explore regulatory frameworks to mitigate the risks of autonomous attack pipelines.
Sources
- Cointelegraph: Google says hackers use AI agents to find and exploit software flaws
- The Defiant: Google Says AI Automated the Jump From Patch to Exploit
Not financial advice. Crypto assets are volatile and you can lose money. Read our Financial Risk Disclosure.
Crypto news checked against two or more sources, prices and a daily market brief.
- Injective publishes new whitepaper days after launch of first US INJ ETF
- Paytaca reports two backend attacks that generated fraudulent BCH orders
- Cypherpunk mines 8,967 ZEC in 44 days, treasury reaches 342k tokens