Hacker
A single human operator can now aim artificial intelligence at your passwords and let the software do the stealing Egor Komarov/Unsplash

A financially driven hacker needed less than six hours to steal thousands of logins after handing almost the entire break-in to artificial intelligence, Google's threat researchers confirmed, exposing how attacks that once took skilled humans weeks now finish before lunch, with alarming implications for cybersecurity teams.

Six Hours From Prompt to Credential Harvesting

Google Threat Intelligence Group, the company's in-house unit tracking state and criminal hackers, detailed the case in its September AI Threat Tracker.

After breaking into a victim's cloud infrastructure, a suspected financially motivated actor combined an AI coding chatbot, a single written prompt, and a library of preset instructions to build a credential-harvesting pipeline that then ran largely on its own.

It scanned for weaknesses, collected credentials, and rotated through the victim's own network addresses so the traffic looked legitimate. The operation harvested thousands of third-party credentials, without a human steering each step.

The Human Google Still Found in the Loop

There's an important catch. Google says it has not yet seen a fully autonomous attack pipeline turned loose on real targets. A person still wrote the prompt and the agent instructions that set the machine in motion. The company's Mandiant investigators traced the second-quarter 2026 break-in and watched the framework handle the drudge work humans used to grind through by hand.

The shift isn't that AI replaced the hacker. It's that one operator can now direct an attack that behaves like a team, without needing an entire specialist crew.

Google also tracked a China-linked espionage group experimenting with automated exploitation tools. That effort involved human-directed activity, and Google disabled assets associated with the group's AI misuse. The September case was highly automated, not fully hands-off.

From AI-Enabled Malware To AI Zero-Days

The credential raid follows a fast escalation. In November 2025, Google identified malware families named PROMPTFLUX and PROMPTSTEAL that use large language models during execution to rewrite code or generate malicious commands. PROMPTSTEAL was already tied to Russian military hackers targeting Ukraine.

By May 2026, researchers reported something sharper still. Google said it had high confidence criminals used AI to help discover and weaponise a previously unknown flaw, a zero-day that bypassed two-factor authentication on a widely used open-source tool.

John Hultquist, chief analyst at Google Threat Intelligence Group, called it 'the tip of the iceberg' and warned that other AI-built zero-days are probably already out there. Google worked with the vendor to address the flaw and disrupt the planned mass exploitation.

Why Your Passwords Are Now the Target

Strip away the jargon and the target is your inbox, your bank, and every account chained to them. Credential theft can lead to account takeovers, drained balances, and ransomware, and AI is pushing its speed and scale up hard.

Faster attack preparation means defenders have less time to notice and slam the door, particularly when automated operations overwhelm conventional monitoring. Security teams are urged to watch for suspicious activity, lock down access tokens, and treat stolen logins as the first domino.

Google's warning is blunt. The era of adversaries using AI isn't coming. It's already here.