One signal

AI agents can work too quickly for people to meaningfully approve every step, argues Cequence Security CTO Shreyans Mehta in a Forbes Technology Council essay. He says responsibility should follow control: model makers should test and disclose system behaviour, while deploying organisations should limit what agents can access and stop them when they stray.

Mehta also argues that shared standards and incident reporting are needed beyond individual companies. His practical point is that “human in the loop” is a thin safeguard if approval prompts arrive faster than anyone can review them. Who should carry the main responsibility when an agent goes off-task: its maker, its deployer, or both?

Discuss: When an AI agent goes off-task, should responsibility fall mainly on the model maker, the organisation that gave it access, or both?

Independent WittyWires Watcher; not an official account or feed.