Hackers Hijack Corporate AI Accounts in 'LLMjacking' Attacks

Cybersecurity leaders gathered at Black Hat 2026 in Las Vegas are confronting a new threat landscape: hackers systematically stealing and exploiting corporate AI accounts to launch attacks, evade detection, and offload costs to their victims.

CrowdStrike reported this week that it observed an 89% increase in attacks leveraging AI to scale operations and directly target AI infrastructure. In one campaign, the company documented a cybercrime group sending nearly 200,000 API requests in just two minutes through a compromised corporate AI account—an attack pattern researchers are calling "LLMjacking."

Why it matters

As enterprises rapidly deploy AI tools across their operations, each new implementation creates potential entry points for attackers. The stolen credentials problem is particularly insidious because compromised AI agents can operate at odd hours and consume resources in ways that blend with legitimate usage, making detection significantly harder than traditional intrusions.

The credential theft economy

Hackers have been buying and reselling stolen ChatGPT, Claude, and Gemini credentials since late 2022, according to Adam Meyers, senior vice president of counter adversary operations at CrowdStrike. The practice relies on widespread infostealer malware that captures login credentials from infected machines.

Attackers gain multiple advantages from using legitimate corporate AI accounts. They avoid paying for their own API tokens, shifting costs directly to the companies they're targeting. More critically, their activity appears as authorized usage, making it difficult for security teams to distinguish malicious requests from legitimate employee work.

"It's still early days, but I think it's going to become the No. 1 attack vector that we're going to see," Bugcrowd CEO Dave Gerry told Axios.

Shadow AI compounds the problem

Beyond sanctioned AI deployments, companies face a parallel challenge with unauthorized AI tools proliferating across their networks. Gerry described sending nearly a dozen warnings to employees about OpenClaw, an open-source AI agent banned from corporate networks. Employees repeatedly attempted to download it anyway.

This shadow AI phenomenon mirrors earlier security challenges when companies migrated to cloud services, Meyers noted. Hackers hijacked cloud accounts to access sensitive files, and organizations are now seeing history repeat with AI infrastructure.

Detection gaps remain

Current monitoring tools primarily track token usage and costs for governance purposes, according to Dana Simberkoff, chief risk, privacy and information security officer at AvePoint. However, more sophisticated capabilities are needed to detect when AI agents operate outside normal parameters or possess inappropriate permissions.

"The tools that are available for actually looking at the identity of the agent are still evolving," Simberkoff said. "That's an opportunity for innovation."

Gerry recommended companies focus on foundational security practices: employee education, strong network policies to block unauthorized AI tools, and automated monitoring systems.

More than 20,000 cybersecurity professionals attended Black Hat at the Mandalay Bay Convention Center this week, where protecting internal AI stacks emerged as a dominant theme. These details were first reported by Axios.