Category: AI Agent Security | Reading time: 3 min
Warp, the terminal-turned "agentic development environment" that says it is used by more than 800,000 developers, has disclosed a breach that reached its users' AI agent logs. CEO Zach Lloyd described the incident in a post on X on October 26, which Warp reposted from its own account.
What Warp has said
According to Lloyd, Warp became aware on Friday, October 23 of "an active and narrow-banded exploitation of a vendor in Warp's infrastructure" and "detected the operation within hours." The attackers ultimately accessed "a limited amount of agent trace and telemetry data," meaning records of what users' coding agents were asked to do and what they did.
Lloyd put the upper bound of affected users at roughly 1%. Warp says it has notified the users it has positively identified, engaged CrowdStrike to investigate, and informed law enforcement.
Lloyd also said the attack "appears to have come from an agent swarm using a very sophisticated penetration and escalation technique, likely starting with a hole in a vendor dependency." That is Warp's early assessment and has not been independently confirmed. Warp has not named the vendor or the flaw, and has not said exactly which fields the stolen logs contained.
Second AI coding tool breach in ten days
The disclosure follows the OpenCode incident of October 17-18, in which stored AI provider API keys belonging to about 1.4% of OpenCode Console users were exposed and some users saw unauthorized charges above $1,700.
Why it matters
Agent activity logs are becoming a target in their own right. A trace of what an agent did can contain the instructions it was given and the systems it touched. Organizations deploying agents should know what their agent records hold, where those records live, which vendors can reach them, and whether they capture the decision trail without carrying the full payload.
Source: BleepingComputer, TSN, TechCrunch on OpenCode