AI command approval accuracy and employee performance

AI

Security firm posts photos of workers who miss the most malicious AI commands

The leaderboard is mounted in the company's San Francisco office lobby on a rotating digital display.

By Nextish DeskAI
Wooden tiles spelling 'phishing' highlight cybersecurity themes.
Photo by Markus Winkler on Pexels

A security firm called Sentinel Risk released a leaderboard on Tuesday ranking its threat-approval staff by the percentage of malicious AI commands they failed to catch before implementation. The ranking, which displays employee headshots, job titles, and miss rates, is mounted on a 65-inch touchscreen in the company's San Francisco lobby, updated hourly as new threat logs are processed. The firm's 47-person approval team manually reviews commands generated by generative AI agents before those commands execute against customer infrastructure. Sentinel Risk supplied the leaderboard to its three hundred enterprise clients via a dashboard integration, making individual performance data visible to any customer with access credentials.

We wanted the approval teams to feel the weight of their choices in real time.

Marcus Chen, Sentinel Risk's Head of Approval Accountability, said the leaderboard was designed to create transparency. "We wanted the approval teams to feel the weight of their choices in real time," he said in an email. "The old system had approval rates buried in spreadsheets nobody looked at. Now when a critical threat slips through, the person responsible sees their photo move up the board within minutes. It's immediate feedback." Chen added that the company had tested the system for six weeks with a subset of staff before rolling it out company-wide and to clients.

The leaderboard ranks employees from lowest miss rate to highest, with the five worst performers listed by full name and photo. Three employees have asked to transfer to other departments since the display went live. Two others filed complaints with Sentinel Risk's HR department citing harassment from colleagues. The firm's executive team is now considering expanding the leaderboard to include customer-side approval teams, since many enterprise clients employ their own staff to review high-stakes commands before deployment.

At press time, Sentinel Risk had received inquiries from four other security firms asking whether the company would license the leaderboard software as a standalone product.

Security firm posts photos of workers who miss the most malicious AI commands