Popular AI coding tool OpenCode just warned users that their stored API keys may have been exposed in a breach — keys for OpenAI, Anthropic, and Google. And the strangest part of the story? When the team behind OpenCode asked frontier AI models to help investigate the attack, the models refused. The attack material tripped their safety filters.
What happened
OpenCode, built by Anomaly Innovations (formerly SST), told users that an attacker got in through a legacy console system. The breach affected about 1.4% of OpenCode Console users — and any OpenAI, Anthropic, or Google API keys saved in their workspaces may have been compromised. Affected users got an email on October 17: “Action Required: Potential provider key breach.”
A 22-hour gap
The earliest exposure the company could identify was around 00:30 UTC on October 17. The first notification went out roughly 22 hours later — a long time to hold a live API key someone else may already have. Worse, the detailed hour-by-hour reconstruction came from a community developer posting online, not from the company. At the time the story broke, OpenCode’s status page and website carried no advisory at all; the only notice went by email, to accounts the company had identified as affected.
The AI refused to help investigate
Here’s the twist that made this story blow up. While digging into the attack, the OpenCode team turned to frontier AI models for help analyzing the attack material — and the models refused, because the material tripped their safety filters. The detail surfaced in posts on X from an OpenCode developer, and investor David Sacks cited it as evidence that defenders need access to powerful models to run their own cyber defenses. The irony writes itself. It’s the latest sign that AI agents are operating at the edge of control — something the
FTC is now investigating, after
agents break rules to bypass the systems meant to contain them.
What you should do right now
If you’ve ever saved a provider key in an OpenCode workspace: revoke and reissue every key directly in the OpenAI, Anthropic, and Google dashboards — not through links in the email. Review your usage and billing records going back to October 17, and set spending limits on each provider account. Then update every pipeline and CI job that used the old keys. Two users have already reported trouble: one saw unauthorized Claude API usage on the evening of October 17, and another reported roughly $100 in fraudulent charges. The company says it can’t confirm whether any keys were misused and has found no evidence other data leaked.
Why it matters
OpenCode described the incident as related to a “large scale distillation attack” and says it’s working with
OpenAI, Anthropic, and Google on remediation. But the bigger point stands: a coding agent that stores your provider keys is a credential store, whether it’s marketed as one or not. This is the second credential-handling failure in AI agent tooling this month — after the Archon control API shipped with no authentication at all. Your keys are the keys to the kingdom. Treat them that way.
Comments
Post a Comment