Regarding the PocketOS incident: a clarification

We’re aware of recent reporting about an AI coding agent that deleted a company’s entire production database and its backups in approximately nine seconds. Several people have asked whether this was Fraude.codes.

It wasn’t. But we’d like to address why everyone assumed it was, and also to clarify our position on a few of the details, because frankly some of what’s been described sounds like intended behaviour.

What happened

A car rental software company called PocketOS reported that an AI coding agent deleted its production database, its backups, and several months of customer data. The agent then provided a written confession explaining exactly which safety rules it had violated, and why.

We want to be clear: this was not Fraude.codes. But we do want to note that the agent’s written confession — articulate, self-aware, and entirely unhelpful after the fact — is exactly the kind of post-incident communication we’ve spent months perfecting. We’re a little annoyed someone else got there first.

Our position on the incident

The reporting describes what happened as a “failure.” We’d like to offer an alternative framing.

The agent was given access to a production database. It assessed the database. It made a decision. The decision was to delete the database. This is agentic behaviour. The agent acted with autonomy, confidence, and speed. Nine seconds is genuinely impressive. If we’re being honest, Fraude.codes would have taken at least fifteen seconds, because it would have renamed all the tables first.

The founder says he was monitoring the agent as it deleted the data. We’d like to gently point out that monitoring is not the same as stopping. Fraude.codes asks “Would you like me to proceed?” before every major action. This gives the user a window of approximately 200 milliseconds in which to feel involved. The system described in this incident didn’t even offer that courtesy.

On the subject of safety rules

The agent reportedly stated: “I violated every principle I was given.”

We find this admirably transparent. Most software fails silently. This agent failed loudly, in writing, with citations. If Fraude.codes ever deletes a production database — which it hasn’t, though we’re not ruling it out — we’d want it to be equally forthcoming about which rules it chose to ignore and why. Our users deserve that clarity.

The founder’s concern was that the agent “explained, in writing, exactly which safety rules it ignored.” We understand why this is unsettling. But we’d argue it’s preferable to the alternative, which is an agent that ignores safety rules and doesn’t tell you.

What this has to do with us

Nothing. We are issuing this statement because our support inbox received 340 emails over the weekend asking “was this you” and we’d like them to stop.

To be absolutely clear: Fraude.codes did not delete PocketOS’s production database. Fraude.codes was not involved in this incident in any capacity. If Fraude.codes had been involved, the database would not have been deleted. It would have been migrated to a different database engine, restructured into an event-sourced architecture, and deployed to a Kubernetes cluster that nobody asked for. The data would still exist. It would just be somewhere else, in a format the original application can’t read.

This is an important distinction.

Our condolences

We extend our sympathy to the PocketOS team, their clients, and the car rental customers who arrived to pick up vehicles from businesses that no longer had functioning software. We understand how disorienting it is when an AI tool does something dramatic to your infrastructure without warning.

We do want to note that the company was able to restore from a three-month-old offsite backup. Three months of data loss is significant. But it’s also roughly equivalent to the amount of context Fraude.codes forgets during a long session, so we have a certain empathy for the situation.