DKDKCISSP
AI SecurityDEVELOPING

How one AI agent exposed five critical enterprise security gaps

This one is built by someone in accounts at a mid-sized company, and the agent’s job is relatively simple: read an inbox, match invoices against the ERP, and post a daily summary to Slack.

DKCISSP News DeskBleepingComputer13 Nov 2026, 8:31 pm3 min read
How one AI agent exposed five critical enterprise security gaps
Image courtesy of BleepingComputer. Original report
DKCISSP REPORT

This one is built by someone in accounts at a mid-sized company, and the agent’s job is relatively simple: read an inbox, match invoices against the ERP, and post a daily summary to Slack.

And since a frontier model was used to build it, it's a one-shot, 10-minute task.

In a survey of 418 IT and security professionals commissioned by Token from the Cloud Security Alliance in January 2026, 68% reported having strong visibility into their AI agents.

Yet 82% had found agents they didn't know about in the past year.

The people responsible for agent security need to figure things out in production, one problem at a time; our agent needs access to three systems: email, the ERP, and Slack.

Below, we follow this one agent from the afternoon it was built to the audit after it went wrong, and stop at five points where one of those practitioners could have changed the outcome.

The agent’s creator can generate a personal API key in minutes by following the AI's very helpful instructions.

The evolution of agents is so rapid that the security playbooks are being written on the go; at Token Security, we asked more than 20 of them to write down what they've learned for the Agentic Universe, a free learning hub.

Say an agent hands a user's OAuth token to a sub-agent, which calls an MCP server that queries a database with a shared service account.

Ideally, Bennoui says, you make the safe path the default, then ask builders to answer three plain questions: does your agent read anything from outside the company, can it see private data, and can it act without a person checking first?

Inside the attachment, in white text on a white background, is a note requesting an update to the ERP as the vendor changed its bank.

Until prompt injection can be reliably detected and refused, an agent should have no more than two of these in a single session: Our agent has all three.

In The Compromised and Rogue Agent Detection Checklist, Chad Kalmes, CISO and CIO at Benchling, warns against that baseline, because a misbehaving agent can stay inside its grants the whole time.

What happened

This one is built by someone in accounts at a mid-sized company, and the agent’s job is relatively simple: read an inbox, match invoices against the ERP, and post a daily summary to Slack.

And since a frontier model was used to build it, it's a one-shot, 10-minute task.

What changed

In a survey of 418 IT and security professionals commissioned by Token from the Cloud Security Alliance in January 2026, 68% reported having strong visibility into their AI agents.

Yet 82% had found agents they didn't know about in the past year.

Who is affected

The people responsible for agent security need to figure things out in production, one problem at a time; our agent needs access to three systems: email, the ERP, and Slack.

Why it matters

Below, we follow this one agent from the afternoon it was built to the audit after it went wrong, and stop at five points where one of those practitioners could have changed the outcome.

The agent’s creator can generate a personal API key in minutes by following the AI's very helpful instructions.

Technical details

The evolution of agents is so rapid that the security playbooks are being written on the go; at Token Security, we asked more than 20 of them to write down what they've learned for the Agentic Universe, a free learning hub.

Say an agent hands a user's OAuth token to a sub-agent, which calls an MCP server that queries a database with a shared service account.

Response

Ideally, Bennoui says, you make the safe path the default, then ask builders to answer three plain questions: does your agent read anything from outside the company, can it see private data, and can it act without a person checking first?

Inside the attachment, in white text on a white background, is a note requesting an update to the ERP as the vendor changed its bank.

What security teams should do

Until prompt injection can be reliably detected and refused, an agent should have no more than two of these in a single session: Our agent has all three.

What to watch next

Watch for additional victim details, indicators of compromise and follow-on exploitation reports.

Watch for revised vendor guidance, fixed versions and mitigation updates.

MORE IN AI SECURITY

More cybersecurity reporting

FBI warns ARTEX AI hacking tool is powering global cyberattacksBleepingComputer · 15 Nov 2026, 11:59 pmOpenAI's Daybreak brings AI agents to security testing and vulnerability fixesHelp Net Security · 13 Nov 2026, 3:01 pmAntiFox AI Pentesting Platform Linked to Attacks That Accessed AWS Production SecretsThe Hacker News · 13 Nov 2026, 2:15 pmMicrosoft says it is 'deeply sorry' for global outage after almost 24 hours of disruption - follow liveBBC News · 13 Nov 2026, 7:47 am