Botpress breach exposes customer AI keys and CRM credentials

An attacker exploited a Botpress internal proxy to steal a staff credential and access customer bot configs containing third-party API keys.

Tools ยท Source: Botpress

What happened

On November 6, Botpress disclosed a major security breach affecting customer workspaces. An attacker began probing a flawed internal authentication proxy on October 12. This proxy sat in front of several internal tools used by the Botpress team. The attacker used the flaw to forge sessions that looked exactly like legitimate Botpress staff accounts. They spent over a week mapping the internal systems and understanding how the platform worked. By October 21, the intruder successfully stole an employee API credential that had administrator access to multiple customer environments.

The attacker used this compromised employee credential to read bot configurations across 16 specific customer workspaces. These configurations are highly sensitive because they store the secrets needed to run integrations. They contain AI provider API keys, messaging channel tokens, and CRM credentials. The attacker also exploited weaknesses in the setup flow of some integrations. They used these weaknesses to alter integration settings in two additional customer workspaces. The breach gave the attacker direct visibility into customer account information and internal staff data.

Botpress logs cannot determine exactly which bots held valid credentials during the specific time of the intrusion. The company is now asking affected workspace owners to treat all stored credentials as completely exposed. The attacker operated a large network of accounts on the platform to consume AI credits. Everything points to a single goal of obtaining AI usage to resell on AI black markets. Botpress fixed the proxy vulnerability on October 21, revoked the compromised employee credential, and blocked the attacker from the network.

Key facts

Why it matters

If you build with AI platforms, your software supply chain is under active attack right now. Hackers are no longer just going after your main servers or customer databases. They are specifically targeting the low-hanging fruit sitting in your third-party chatbot configurations. A single compromised staff account at your vendor can easily leak your most sensitive OpenAI keys and CRM access tokens. This means your startup infrastructure is only as secure as the internal tools used by your platform providers. You carry all the risk when their internal security fails.

This incident fundamentally shifts how founders must manage API secrets moving forward. You can no longer trust a platform to perfectly guard your keys just because they have a good reputation or a slick interface. Builders will need to implement strict usage limits and rapid key rotation policies by default. Expect underground AI black markets to drive a massive increase in attacks specifically aimed at developer tools and agent platforms. The financial demand for stolen AI compute is simply too high for attackers to ignore these vulnerabilities.

For builders

Rotate exposed third-party keys immediately

If you use Botpress, check your workspace integrations right now. Rotate any AI provider keys, messaging tokens, or CRM credentials stored in your bot configurations. You pay the ultimate price if attackers drain your AI credits or access your customer data.

Enforce strict API billing limits

Attackers want to steal your keys to resell AI usage on black markets. Set hard billing limits on your OpenAI and Anthropic accounts today. This simple step stops a leaked key from bankrupting your startup overnight while you sleep.

Audit vendor staff access policies

The attacker used a Botpress employee credential that had standing admin rights to customer workspaces. Ask your vendors exactly who has access to your production environments and why. Push your platforms to adopt zero-trust architectures for their own staff to protect your data.

My take

AI tools are the new prime target for hackers looking to make a quick profit. I always tell founders you cannot blindly trust a platform to guard your keys when their own internal proxies are vulnerable to basic probing. Build your apps assuming every third-party secret will eventually leak, and design your systems to survive that exact scenario.

Original reporting: Botpress. This is my rewrite and opinion.

More AI news for builders