OpenAI refunds API credits drawn in the OpenCode Console attack
OpenAI says credits drawn from keys saved in OpenCode Console were refunded. Still delete the key, set a hard spend limit, and apply for Daybreak Blue early.
If you saved an OpenAI API key in an OpenCode Console workspace, OpenAI says credits drawn in the attack were refunded. Still replace that key and set a hard spend limit. Security teams can apply for Daybreak Blue, OpenAI's lower-refusal access for approved defensive work.
What changed#
On Oct 17, OpenCode posted that a breach "impacted 1.4% of OpenCode Console users", and that "it was a subset of users who had external providers linked". Co-founder Dax Raad (his role per The Hacker News) added: "note this is the OpenCode Console, not OpenCode itself, those keys never come to us".
At 12:12 AM ET today, Tibo Sottiaux, who works on Codex and ChatGPT at OpenAI, posted: "Our team worked with OpenCode to have their users who are also OpenAI API customers back in business and any credits refunded which would have been drawn in the attack." He added: "Also, you can request Daybreak Blue access to use our cyber permissive setup for defense. Do not wait until you have to deal with an incident."
Why it matters#
Our read: refunds cover this incident, not the next. Raad said the models OpenCode used to investigate "did quite a bit and then started refusing", and "we got access to daybreak blue now so it's better". He did not name the vendor. OpenAI's help article says "Approval is not automatic", so apply early.
How to use it#
- OpenAI's security guide says: "Delete the affected key through the API key dashboard." Then make a new one.
- Per the same guide, "Review your API usage for unexpected activity and keep the details", then contact OpenAI Support. The Hacker News says to check back to Oct 16.
- Per OpenAI's spend limits docs, in Project settings, select Limits, then Edit spend limit under Spend. Enter the Monthly spend limit, turn on Enforce a hard limit, and select Save.
- Security teams: individuals go to chatgpt.com/cyber (Tibo's link) and select Set up Daybreak; organizations use the request link in the help article.
Prompt (ours, untested), for an approved team in Codex with the Daybreak toggle on:
Here are our gateway access logs for the last 48 hours. Group requests by API key and source IP, flag keys used from new IPs or at unusual volume, and rank which to revoke first, citing the log lines.
Good to know#
- Refund details. Tibo gave no counts or amounts, and did not say whether you must ask.
- Daybreak Blue limits. Per the help article, Astra and GPT-6.1 Sol keep standard safeguards under Blue, and individuals need "an eligible paid plan, Advanced Account Security, and at least one compatible FIDO2 hardware security key."
- Watch next: OpenCode's root-cause report, and whether OpenAI says how refunds are applied.