In June 2026, the competitive intelligence tool Klue was breached. According to the timeline compiled by security firm Huntress, the attacker first got into Klue with a long-unused but still active credential that Klue had created years earlier for a prototype integration. The attacker then planted code in Klue’s systems to collect customers’ OAuth tokens for platforms such as Salesforce and Gong, and used those tokens to query customers’ CRMs directly and carry the data away. These tokens are passes that let a third-party tool access data on your behalf. Huntress itself was one of the victims; what was taken included business contacts, products trialed, subscription pricing, and sales correspondence.
This entry point wasn’t at the victim companies. The old credential used as a springboard was at the vendor, and no amount of hardening their own logins would have stopped it. But how much each victim lost was decided by one thing: how much the token they had granted Klue could read.
So when a third-party tool connects to your CRM, manage it as a side door. You can’t fix the vendor’s vulnerabilities; you can control three things: how much read access it gets, whether unusual queries are visible, and how quickly you can revoke access in an incident.
Once the token is taken, every query looks legitimate
An integration usually asks “allow access?” once at install time. After you agree, it keeps reading data through legitimate channels. Once attackers hold the token, their queries also look like legitimate API calls; they don’t need to break into the CRM itself.
Security firm ReliaQuest’s analysis found that the attackers first listed every data object in the CRM, then ran large paginated queries; in at least one environment they sent nearly 1,000 queries within 15 minutes, and in another case extraction continued for more than six hours. If someone had been watching, this activity was visible.
Answer three questions for every integration
What can it read? List every third-party tool that can read the CRM and write down which objects it can read: contacts, opportunities, quotes, contracts, attachments, support records, and whether it can write. A competitive analysis tool may need only a few opportunity fields yet hold read access to entire accounts. Narrow wherever possible, and where you can separate environments or objects, don’t share one key.
Who sees anomalies? Set notifications for bulk queries, unfamiliar source addresses, and large after-hours exports, and name who receives them. ReliaQuest also recommends restricting which addresses third-party integration accounts can connect from.
How do you shut it off in an incident? Document each integration’s owner, the steps to revoke authorization and end sessions, and which workflows break when it’s disabled and who takes over. Huntress recommends revoking all sessions for the affected services, not just the integration.
Start with an integration list
A small team doesn’t need to turn this into a security project at once. Start with an integration list: tool name, owner, what it can read, whether it’s authorized through a personal or service account, how to disable it, and the date last checked. Review it monthly, and disable authorizations no one has managed for 90 days or that have no findable owner.
The list should answer three questions the moment something happens: who’s responsible, what it can read, and how to shut it off. The owner on the list is the person who can decide whether to disable it and whether to notify customers, not necessarily whoever installed it.
Narrower permissions only limit the damage
Narrowing permissions only reduces losses; it can’t keep a vendor from being breached. Once data has been taken, notifying customers, legal judgments, and insurance claims are decisions for the company’s responsible people, not something engineers or AI can handle alone. Huntress also warns against opening leaked data archives yourself, since such files often contain malware; leave verification to your incident response team or legal counsel.
The same logic applies to AI workflow tools inside your company: which keys the flows hold determines the scope of an incident; see When an AI Workflow Tool Is Exposed, First Find Which Keys Its Flows Hold. For giving a cross-system AI agent permissions step by step, see When an AI Agent Works Across Systems, It Needs a New Key at Every Stop.
AI handoff card
I want to inventory which third-party tools are connected to our CRM, to understand how large our losses could be if one of them is breached and how quickly we could stop the bleeding.
The method comes from this article: https://boosterminiclass.com/en/posts/crm-integration-token-breach-checklist/
Its view: when a third-party tool is breached, the CRM loss depends on the access granted to it. You can't fix the vendor's vulnerabilities; you can control how much read access it gets, whether unusual queries are visible, and how quickly you can revoke in an incident.
I'll provide a list of integrations and their currently known authorization scopes (without any keys, tokens, or passwords). Please:
1. Organize them into an integration list: tool name, owner, which objects it can read, whether it can write, the account type used for authorization, and how to disable it; mark missing fields as "to confirm";
2. Flag integrations whose permissions clearly exceed their purpose, and suggest how to narrow them;
3. Recommend notifications for unusual queries (such as bulk queries in a short time, unfamiliar sources, after-hours exports) and who should receive them;
4. Draft incident revocation steps, including which authorizations to revoke, which sessions to end, which workflows will break, and who takes over.
For whether to notify customers or regulators, only list the questions the owner must decide; don't draw conclusions for me.
References
- Huntress blog: Cybercrime Breaches Klue: Salesforce Data Impacted for Many Victims, including Huntress — https://www.huntress.com/blog/klue-breach-investigation [accessed: 2026-10-11]
- ReliaQuest: Klue Integration Abused in Salesforce Data Theft — https://reliaquest.com/blog/threat-spotlight-integration-abused-in-crm-data-theft/ (2026-06-17)
- Help Net Security: Klue breach lead to Salesforce data theft, Huntress affected — https://www.helpnetsecurity.com/2026/06/19/klue-salesforce-data-breach-huntress/ (2026-06-19)
