THE HOOK
A coding-agent gateway that held customers' AI keys just learned the hard way that a key vault is a very tempting target.
THE GIST
OpenCode disclosed a breach on 17 October that hit 1.4% of its hosted Console users, the ones who'd linked their own OpenAI, Anthropic or Google keys. Attackers reached a legacy system it was retiring, and some users report drained balances. Anthropic says it has revoked the Claude keys it believes were exposed.
THE TAKE
Storing other people's API keys so a team never sees them is convenient right up until it isn't. A pile of live frontier-model keys is basically a prepaid card for anyone running a distillation farm. "A legacy system we were decommissioning" is the oldest line in breach notices. Retire it first, then brag about the gateway.
OPINION
OpenCode hasn't said how attackers got in or how the keys were stored, and the 9to5AI breakdown is clear that the 1.4% figure doesn't tell you how many keys actually walked out. If you linked a provider key, revoke it at the provider, set a hard spending cap, and treat any hosted gateway that holds your keys as part of your attack surface.
- Story type: Breaking news
- Impact: Medium
- Confidence: High (official announcement)
- Audience: Developers
- What changed: OpenCode confirmed a breach of a legacy Console system that exposed stored provider API keys for 1.4% of Console users.
- What to watch: Whether OpenCode publishes a post-mortem naming the entry point and how many keys were taken, and whether OpenAI and Google follow Anthropic in revoking affected keys.
- Reading time: 1 minute
