We have been dealing with a security attack since yesterday. Our team has been all hands on deck since the start, and we took a large portion of our services offline to prioritize user security.
Our service is progressively coming back online (most of it already is), and we are actively monitoring the situation.
We’ll continue to actively keep our users updated through emails, Slack, and our status page as we monitor the situation and learn more.
With the spate of attacks on infra providers over the past few weeks, it's been interesting to see how different competitors respond. The confidence some of the founders seem to have that they won't/can't be next is telling.
> GitBook has been the target of a sophisticated, highly targeted attack using AI agents
Serious question: How do you know it was using AI agents? What logs distinguish it from a human?
I see this claim a lot and while I don't doubt AI is being used for this, I have a hard time believing every breach has a smoking gun showing it to be an AI assisted attack.
My guess is the speed of exfiltration/exploration of the environment is probably impossible for a human, or even a small team. But I'm also curious what their evidence is.
I’m genuinely sorry the end of friendco.de impacted you. It was 12-13 years ago; the service had little usage at the time, and we probably didn’t handle the end perfectly. I don’t recall exactly, but I can assure you that the team today is more experienced.
Is this what they mean when they talk about accountability laundering?
Oh, you see, we had the best security ever, but they used AI agents, and there's no defence for that, you can't secure a system against those spooky AI agents.
This is not what we meant. We will be accountable and will post a complete post-mortem. But it’s important to understand that the attack was a lot more developed than what we’ve been used to with humans, especially in terms of pace and adaptability.
Note that the summary at the top of https://www.gitbookstatus.com at time of writing does not mention the unauthorized access and the incident details does not mention the contents disclosed above.
GitBook’s CTO here
We have been dealing with a security attack since yesterday. Our team has been all hands on deck since the start, and we took a large portion of our services offline to prioritize user security.
Our service is progressively coming back online (most of it already is), and we are actively monitoring the situation.
We’ll continue to actively keep our users updated through emails, Slack, and our status page as we monitor the situation and learn more.
Nice to see someone not fucking up outreach during a security incident.
Good work.
Good luck... this isn't easy, so best of luck to you and the team.
Classy response.
With the spate of attacks on infra providers over the past few weeks, it's been interesting to see how different competitors respond. The confidence some of the founders seem to have that they won't/can't be next is telling.
After turning on 2FA, the page doesn't update to say it is turned on. It still has a configure button. Reloading the page fixes this.
> GitBook has been the target of a sophisticated, highly targeted attack using AI agents
Serious question: How do you know it was using AI agents? What logs distinguish it from a human?
I see this claim a lot and while I don't doubt AI is being used for this, I have a hard time believing every breach has a smoking gun showing it to be an AI assisted attack.
My guess is the speed of exfiltration/exploration of the environment is probably impossible for a human, or even a small team. But I'm also curious what their evidence is.
Maybe there was a level-of-stupidity tip-off, like trying to break in and access something that's already public...
https://www.gitbook.com/status/security-incident-2026-11-03
Does a security incident page really need to connect to 11 3rd party domains. smh
It's bizarre they have both this and gitbookstatus.com and they contain different info.
We summarized more information on this page from recent learnings, than what has been posted as updates on the status page (https://www.gitbookstatus.com/incidents/01M6FXDDGQRWHMYYWXY2...).
Good that they rotated all the tokens themselves instead of just telling users to do it, seen too many incidents where that step gets skipped.
Mentioned this 5h ago, but this post definitely fleshes out more details.
https://news.ycombinator.com/item?id=50351061
FYI, Gitbook was made by the team behing friendco.de, which they killed without any announcement years ago.
I’m genuinely sorry the end of friendco.de impacted you. It was 12-13 years ago; the service had little usage at the time, and we probably didn’t handle the end perfectly. I don’t recall exactly, but I can assure you that the team today is more experienced.
Not really relevant here.
"GitBook has been the target of a sophisticated, highly targeted attack using AI agents."
Add it to the list, I guess.
Is this what they mean when they talk about accountability laundering?
Oh, you see, we had the best security ever, but they used AI agents, and there's no defence for that, you can't secure a system against those spooky AI agents.
This is not what we meant. We will be accountable and will post a complete post-mortem. But it’s important to understand that the attack was a lot more developed than what we’ve been used to with humans, especially in terms of pace and adaptability.
Oof that formatting is rougher than I thought. Will try to clean it up after I'm done reconfiguring gitbook...
new lines between bullets, only italics available, not markdown
This is the public disclosure: https://www.gitbookstatus.com/incidents/01M6FXDDGQRWHMYYWXY2...
Note that the summary at the top of https://www.gitbookstatus.com at time of writing does not mention the unauthorized access and the incident details does not mention the contents disclosed above.
One thing that would be appreciated are more details in these sorts of reports.
"the attacker gained unauthorized access to some of our infrastructure credentials"
How? Which infrastructure?
We’ll provide details in a long post-mortem. For now, it was important to notify our users as early as possible.