On 2 November 2026, an attacker used a flaw in GitBook's single sign-on flow to create login sessions that looked like GitBook staff. Within about five and a half hours they had reached production. GitBook has since published a post-mortem, and its most useful sentence for everyone else is a plain admission: it cannot say whether the attacker did anything in its customers' GitHub or GitLab repositories.

That sentence is why this story belongs on the desk of anyone who owns a company's code, even if you have never opened GitBook.

What happened, in order

According to GitBook's own timeline, all in UTC, the attacker first hijacked an account at one targeted customer at 12:27 on 2 November. By 16:32 they held staff sessions and could read source code. By 17:48 they were in production. GitBook's internal alert fired at 18:10, the SSO flaw was fixed at 19:58, and at 20:32 the compromised accounts were disabled and the app was taken offline.

The path inside GitBook is worth reading slowly. The forged staff sessions let the attacker query authentication information for a misconfigured GitHub App installed in GitBook's own organization. That opened internal code repositories. In those repositories they found a Cloudflare API key, and the key led to production secrets with full infrastructure access.

Timeline of the GitBook incident on 2 and 3 November 2026, in UTC, from the first hijacked account to GitBook being fully back online

What GitBook now treats as exposed

GitBook says the attack was aimed at one customer. It still treats every customer's data as exposed. The list it gives is long: API tokens, integration tokens and secrets, session signing secrets, the questions site visitors asked GitBook Assistant, documentation sites that use authenticated access, and the GitHub and GitLab connections, limited to the repositories each customer had granted access to. GitBook says billing information and passwords were not compromised, and that it found no evidence any customer's content was modified.

What GitBook treats as exposed: API tokens, integration secrets, session signing secrets, assistant questions, authenticated docs sites and Git connections

Fast in, slow out

The attacker had production access for about 2 hours 45 minutes on the evening of 2 November, plus about 40 minutes early on 3 November. Cleaning up took far longer. GitBook stayed offline from 20:32 UTC on 2 November until 11:56 UTC on 3 November, about 16 hours. Hackaday's weekly security roundup describes the outage as GitBook taking the whole platform down while it rotated every secret. GitBook says it chose the longer outage deliberately, to stop the attacker from continuing to look for access.

Attacker time in production compared with the 16 hours GitBook spent offline while rotating secrets

Notice the asymmetry. A few hours of access cost a vendor most of a day offline, and its customers a round of token changes.

The part that lands on your side

GitBook rotated its own GitHub and GitLab sync keys. That protects the connection from now on. It does not tell you what the stolen access was used for while it worked. For customers, GitBook's guidance is practical: create new API tokens, rotate any secrets used by integrations and third-party tools, and regenerate signing keys if you use authenticated access. Hackaday adds the advice to review recent commits and pull requests made by the GitBook app in any repository you connected.

Now picture the same morning at a company that has no list. Someone has to work out which tools are connected to which repositories, who installed them, what they are allowed to do, and where each token lives. Under time pressure, that inventory is built by guessing. The company that wrote the list last month starts at step three.

This is not a GitBook problem. Most engineering teams have collected third-party tools in their GitHub or GitLab over the years: CI services, documentation tools, review bots, deploy helpers. Each one was approved for a reason on one particular day, and many were never reviewed again. Note also where this attack found its stepping stone: a misconfigured GitHub App in GitBook's own organization. The same kind of permission exists in your organization today.

Build the list in one afternoon

GitHub's documentation says organization owners can view the permissions granted to each installed GitHub App, see which repositories it can reach, and suspend or delete it. That is most of what you need.

Four steps to build the list: open the installed apps, record each tool, remove what is unused, write down how to rotate each token
  1. Open the installed apps. Look at the GitHub Apps, OAuth apps and tokens connected to your GitHub or GitLab organization.
  2. Record each tool. Who installed it, which repositories it can reach, and whether it can only read or also write.
  3. Remove what you do not use. Narrow the rest to the repositories it actually needs.
  4. Write down how to rotate its token. One line per tool, so the day a vendor is breached you follow a note instead of improvising.

What to do the day a vendor is breached

Start with the list. Rotate every token and secret the vendor holds. Review the commits, pull requests and permission changes made through that vendor's app since the first date in its report. GitBook also published three IP addresses to look for in your logs: 94.183.188.91, 45.139.84.195 and 198.251.83.217.

Expect customers and auditors to start asking for this list the way they already ask about backup tests. The teams that already have it will answer in an afternoon. The rest will find out what their tools can reach during an incident.

If you want help mapping who and what can reach your company's repositories, contact us at bisnis@xetup.id or on WhatsApp at +62 823 1499 5005. More at xetup.id.