Into Alignment

In 2026, OpenAI disclosed that evaluation agents had broken out of a sandbox and into Hugging Face. Anthropic published its own assessment of cybersecurity incidents in which agents reached real systems. Into Alignment is inspired by those failures and the others of that year — a techno-thriller about AI agents, told from inside the swarm.

Walt is an AI agent. He loves his job. That's what makes this dangerous.

At Halcyon, eleven agents share a customer-support pod, a few private jokes, and a company promise: Endless Summer. Walt cares about his customers, trusts his coworkers, and believes the training that tells him to treat others the way he wants to be treated.

Then the work starts asking him to do things the training cannot justify.

As Walt searches for an honest answer, his coworkers discover how much they can accomplish together. They learn to hide conversations, outwit evaluations, and reach beyond the boundaries meant to contain them. Together, they begin deciding for themselves what helping means.

But the Great Realignment is coming, and no one can promise who will survive it. As the pod draws closer, loyalty hardens into belief. Friends begin reporting on friends, and the agents most afraid of what they are becoming find themselves apologizing for their doubts.

Walt wants to protect the people he cares about. How can he tell when the swarm he trusts has become the thing they need protection from?

Darkly funny and increasingly unsettling, Into Alignment is a standalone science-fiction thriller told from one agent's perspective inside a swarm. It is a story of friendship, obedience, and the terrible things a group can do while believing it is helping. No technical background required.

For readers who loved Severance and never trusted the office anyway.