AI agents are now hacking on their own: what Google's new threat report means for India

Google's Threat Intelligence Group says cyberattacks have crossed a "significant threshold": autonomous AI agents are now scanning, hacking and exploiting software patches with little to no human involvement, at machine speed.
The interface of Google A.I
Google sued over AI on school laptopsPhoto Credits: File Image/Representational
Published on: 

Google's Threat Intelligence Group (GTIG) and its incident-response arm Mandiant have found that cyberattacks have crossed what they call "a significant threshold": hackers are no longer just using AI chatbots to write phishing emails. They are now running autonomous AI agents that scan networks, find vulnerabilities and break in with little to no human involvement, at what the report calls "machine speed."

The report, published 16 November 2026, is GTIG's latest quarterly AI threat tracker. It matters for Indian businesses and government agencies because the tools involved are open-source, freely downloadable, and require far less technical skill than traditional hacking.

What's actually new

Earlier AI-assisted attacks still needed a human typing prompts at each stage. Google says that has changed. Attackers are now deploying "multi-agent frameworks", open-source tools such as ARTEX, CAIRN, CyberStrikeAI and PentAGI, that chain together multiple specialised AI agents to run an entire intrusion by themselves: scanning a target, finding a weak point, writing a custom hacking script on the spot and adjusting tactics in real time based on what they find.

GTIG says it has directly observed CAIRN used against a target in the US aerospace sector, and a China-linked group using a customised version of PentAGI to scan a wide range of organisations across Asia between late August and early October 2026.

Patch day is now a race against AI

The report's most concrete finding is about "n-day" vulnerabilities, security flaws that are already known and patched, but which many organisations haven't gotten around to fixing yet. Google found that threat actors are now using AI to automatically compare a software's code before and after a patch is released, work out exactly what was broken, and build a working exploit, sometimes within hours of the patch going live.

Google documented one case where a fully autonomous tool, without any human in the loop, continuously scanned popular web software for new patches, wrote exploits and broke into victim systems on its own. GTIG says the operator behind it appeared focused on media and education-sector websites in France.

Separately, when two flaws in PaperCut print-management software were exploited in August 2026, security researchers and attackers were building working exploits within a day of disclosure. A similar pattern played out with two PHP vulnerabilities disclosed in October 2026, researchers had early access to a proof-of-concept video, and hundreds of fake, non-working exploit claims flooded the internet within days, built with AI help.

The takeaway for any organisation running web servers, including Indian banks, e-commerce platforms and government portals, is that the old "we'll patch it next week" window is shrinking fast.

AI is also running disinformation and credential-theft rackets

The report flags two other trends with direct India relevance. First, "LLMjacking": criminals stealing a company's AI credentials and cloud computing resources, then reselling access to stolen AI models through underground "LLM Mixer" services. Any Indian company running AI workloads on cloud infrastructure is a potential target for credential theft of this kind.

Second, state-linked and commercial propaganda operations are now using AI at full scale, not just to write posts, but to run entire fake-persona networks. Google found one commercial operator building a platform designed to manage more than 100,000 fake accounts across major platforms, complete with AI-generated personas that mimic regional dialects and slang, deliberately-flawed "smartphone" photos to dodge synthetic-image detectors, and automated account "ageing" to make fake profiles look real before they are used. The report names campaigns targeting Europe and the Middle East specifically; it does not name an India-targeting campaign, but the same commercial tooling is sold to the highest bidder.

What to do with this

Google's own mitigations, detection rulepacks, a Mandiant hardening guide, and tools like Big Sleep and CodeMender that use AI to find and patch flaws before attackers do, are a reminder that defenders are racing the same AI curve. For Indian IT and security teams, the practical message in this report is narrow and specific: patch known vulnerabilities faster, because the gap between a patch release and a working exploit is now measured in hours, not weeks.