Warp, an AI coding tool, thinks it was hacked by a "swarm" of AI agents

On Friday, October 23, Warp discovered that someone had broken into its infrastructure. The entry point was not its own software but an outside vendor, which the company has not named. In its analysis, Warp refers to an "agent swarm" and a "highly sophisticated" technique of intrusion and privilege escalation.

Warp is a so-called "agentic" development environment: developers entrust AI agents with writing, testing and deploying their code. Its CEO, Zach Lloyd, made the incident public in a post on X, BleepingComputer reported on October 27. According to him, the teams spotted the operation "within hours" and were able to limit its effects.

What could have been taken?

Warp has identified two types of potentially affected data: the decision logs of its agents, meaning the trace of the steps the AI followed, and telemetry data. This detail matters. An agent log can contain code snippets, file paths and commands that were run. The company has not said whether source code, instructions given to the agents ("prompts") or credentials were among the exposed data.

According to Zach Lloyd, "at most about 1% of Warp users" may be affected. The company claims at least 800,000 users on its website, which would put the figure at a few thousand people. Warp has not confirmed this calculation.

An investigation still under way

Warp says it has notified the users it identified as affected, informed law enforcement and brought in CrowdStrike for the investigation, which continues with outside experts. The term "agent swarm" comes from Warp's own analysis: no published technical evidence so far makes it possible to verify that the attack was carried out by AI agents.

There is a recent precedent: the code platform OpenCode, where API keys stored on behalf of its users leaked. Tools that centralize the work of AI agents are becoming targets in their own right, because they concentrate code, access and activity traces.

What to do if you use Warp

Warp says affected users will receive instructions from the company on the steps to take. In the meantime, here are a few common-sense measures:

• monitor your account activity and be wary of unsolicited messages claiming to come from Warp, as BleepingComputer recommends;
• if your agents handled secrets (API keys, access tokens, database passwords), assume they may have appeared in the logs and rotate them as a precaution;
• check the logs of your repositories and cloud services for unusual access since October 23.

What happens next depends on two answers Warp has not yet given: the name of the compromised vendor, and what exactly the exposed logs contained.