What happened
In October 2026, attackers got into OpenCode, a legacy AI platform that was already on its way out — and found what its operators had left behind: live API keys belonging to the platform's users, connecting to the major AI providers. The earliest exposure was traced to October 17; within days, victims were discovering unexplained charges and suspended cloud projects.
What was stolen was not data in the classic sense. What was stolen was the ability to spend. An API key is a credit line: whoever holds it can run up usage on the owner's account. The attackers put the keys to work on industrial-scale AI usage — the pattern of an AI distillation operation, using one provider's models, paid for by strangers, to train competing ones. One account was charged more than $1,700 overnight.
The open door
The system was being decommissioned — but nobody had actually switched it off, and nobody had revoked what it held. The keys stayed live, unreviewed and unmonitored, on a platform that had fallen off the org chart and out of the audit scope. Attackers do not only attack what you run. They attack what you forgot to switch off.
The cost
The direct bill landed on individual users first: overnight charges past $1,700 on a single account, cloud projects suspended as providers reacted, and hours of incident response across every organisation whose keys were on the platform. The deeper cost is structural — an AI key with no spending cap and no alert is an open wallet, and most companies still cannot answer a basic question: where do our AI credentials actually live?
What boards should do
- Know where your AI credentials live — every platform, every key, every owner — and have someone review that inventory on a schedule.
- Put spending caps and real-time alerts on AI accounts, the way you already do on corporate cards. An uncapped key is an uncapped liability.
- When a system is retired, verify that it is truly gone: keys revoked, access paths closed, data deleted — with a name attached to the verification.