International Edition
Latest News
Technology

Google Warns of New Era of Autonomous AI-Driven Cyberattacks

Autonomous AI exploit frameworks arrived on Monday when the Google Threat Intelligence Group and Mandiant published their latest GTIG AI Threat Tracker report, revealing an attack pipeline where artificial intelligence agents find bugs, write exploits, and execute them…

Neat Rolls Out Fix for Brotli Flaw on CISA's Exploited List

Autonomous AI exploit frameworks arrived on Monday when the Google Threat Intelligence Group and Mandiant published their latest GTIG AI Threat Tracker report, revealing an attack pipeline where artificial intelligence agents find bugs, write exploits, and execute them without human sign-off.

Researchers uncovered the framework, designated as exploit_agent in its configuration files, on an exposed staging server, VentureBeat reported. The server also held stolen victim data and post-compromise scripts, though Google did not establish that the framework itself caused those specific breaches.

“In the past quarter, GTIG identified threat actors using LLM frameworks to fully automate vulnerability discovery and exploitation without a human in the loop for the first time,” the UC Today report said.

Google Warns of New Era of Autonomous AI-Driven Cyberattacks
Photo: VentureBeat

Security experts note that the primary shift lies in velocity rather than novel bug creation. “Attackers didn’t invent a new kind of bug. They automated the tedious part: reading the patch, finding what it fixed, and writing the exploit,” said Merritt Baer, former Deputy CISO at AWS and executive advisor to Upwind and G2i, in the VentureBeat coverage.

“A skilled person might spend days on that. Now an agent does it while its operator sleeps,” said Merritt Baer, former Deputy CISO at AWS and executive advisor to Upwind and G2i.

The exploit_agent framework integrates multiple frontier and local large language models to run continuous differential analysis on popular web application software. It monitors software patches, compares code before and after updates to determine fixes, drafts exploits, and executes them automatically.

Configuration files for the framework contained Russian-language comments describing exploit_agent as a pilot and instructing operators to input LLM API keys directly. Telemetry from the server indicated operator interest in French websites across media, entertainment, and education sectors. Google also published a code excerpt showing the framework examining access controls in version 2.19.25 of the Spectra WordPress plugin, though no successful exploit was tied to it.

Beyond exploit_agent, researchers identified several open-source multi-agent offensive frameworks in use by threat actors, including ARTEX, CAIRN, CyberStrikeAI, and PentAGI. Google observed CAIRN targeting the United States aerospace sector.

Automating vulnerability weaponization shortens the time organizations have to apply security updates. Software disclosures and active exploits nearly doubled between January and August 2026, according to Google.

Zero-day exploits climbed from a monthly average of eight in 2025 to 11 per month in 2026, before spiking to 22 in August. After PHP patched two flaws on October 21, hundreds of claimed exploits appeared online, though many failed to operate.

“None of the response is exotic. Know what you expose to the internet, patch it faster than you did last year, and watch for exploitation, not just for vulnerabilities,” Baer told VentureBeat, emphasizing that defenders must rely on internal environmental knowledge to counter attacker velocity.

About the author: Anika Shah - Technology

MSc in Computer Science, senior reporter. Anika focuses on AI ethics, cybersecurity, and emerging hardware—frequently moderating panels at CES and Web Summit. “Anika Shah decodes tech breakthroughs and startup disruption shaping tomorrow’s digital landscape.”