How to Lock Down Your Customer Records as AI Attack Tools Spread
AI attack tools reached customer records at WorkOS and Chainalysis. How a small business locks down its CRM and logins and spots phishing quoting real chats.

Short answer: AI attack tools still break in the old ways: stolen or guessed passwords, apps left open to the internet, and keys nobody changed. A small business can close most of those doors this week. Require two-step sign-in on every tool that holds your customer list, remove accounts nobody uses, and call back before acting on any surprise request.
Key points
WorkOS said on Nov. 2 that an intruder reached an internal app its sales team uses and saw business contact details and summaries of customer conversations. The company said the attacker used an AI-assisted attack framework.
WorkOS founder Michael Grinich named that tool as ARTEX, an open-source AI framework built for penetration testing, BleepingComputer reported on Nov. 3.
Blockchain analytics firm Chainalysis disclosed on Oct. 30 that an outsider using what it called "offensive agentic techniques" reached its internal tools and saw customer names, email addresses, user IDs and role labels.
Security firm CrowdStrike said on Oct. 29 that a single hacking group had aimed ARTEX at more than 200 organizations in at least 14 countries since Oct. 17.
What to do this week: list every app that holds your customer names and emails (CRM, email, invoicing, booking, newsletter) and make sure each one requires two-step sign-in for every user.
What happened at WorkOS and Chainalysis?
Two tech companies disclosed in the past week that attackers using AI tools got into internal business systems and saw customer records. Neither company says its main product was breached.
WorkOS sells sign-in and identity tools to software companies; BleepingComputer notes the company says at least 3,000 companies use it. Its bulletin says it spotted the unauthorized access on Oct. 30. The exposed app held names, emails, phone numbers and job titles from its customer relationship management (CRM) system, plus account histories and summaries and excerpts of sales calls, support talks and customer Slack chats. WorkOS says it has no evidence that its production services or customers' application data were touched. It cut off outside access to the app, hired an outside forensics firm and is reviewing access to its other internal tools.
Chainalysis says it found the intrusion in October. Its notice, also covered by The Hacker News, says it has seen no sign that investigation content, plain-text passwords, MFA secrets or session tokens were reached. In a Nov. 2 update, it said it had taken some products offline as a precaution while outside experts finish their checks.
What is an AI attack framework?
It's software that chains several AI agents together to do an attacker's legwork. In an Oct. 30 report, Google's threat intelligence group and its Mandiant unit described ARTEX as a multi-agent tool that turns a goal into machine commands. Its creator took the original code down in October because of abuse, but copies keep circulating, Google said.
Google says it has seen several attacker groups use ARTEX and similar tools in real break-ins since late August 2026. The agents work in short bursts of about 2 to 25 seconds. In one case, Google logged 27 events in 23 seconds of activity.
Think of a burglar who used to try one door a night. Now he can send a crew to rattle every door on the block in minutes. The doors are the same; the pace is new.
Why should a small business care?
Because these tools go after the same weak spots most small businesses have. Google wrote that the basic mechanics haven't changed, naming compromised credentials and unpatched systems, and that only the speed and scale are new.
Cloud security firm Wiz reported on Oct. 22 that an attacker using ARTEX-linked tooling checked which usernames existed, flooded password-reset pages, signed up for accounts with throwaway email addresses, and ran a large password-spraying campaign. CrowdStrike described a case where the agents used sign-in keys from a 2022 data export that had likely not been changed since.
The stolen data matters too. WorkOS warned that contact details and notes from past conversations can make a fake request look real, and told customers to check unexpected requests "even when a message references a real colleague or support conversation."
Our example (hypothetical): a Kalamazoo landscaping company keeps its customers, quotes and email threads in an online CRM. Someone who gets in could email a customer about "the quote we talked about Tuesday" with new payment details, and the customer would have little reason to doubt it.
How do I lock down the tools that hold my customer list?
Find every copy. Write down each app and file that holds customer names, emails or notes: CRM, email, invoicing, booking, newsletter, shared drives and spreadsheets.
Require two-step sign-in for everyone. The Cybersecurity and Infrastructure Security Agency (CISA) tells small businesses to start with admin accounts and staff who handle sensitive data. It ranks a physical security key as strongest, then an authenticator app with number matching, and text or email codes as the weakest choice.
Remove accounts nobody uses. Close logins for former employees, old contractors and trial users. Give everyone else only the access their job needs, one of the long-standing defenses Google's report lists.
Close doors you don't use. If a customer portal or internal tool doesn't need public sign-up, turn it off, and keep admin pages off the open internet. WorkOS's first fix was cutting outside access to the app.
Change keys after any leak, old ones too. CrowdStrike advises replacing sign-in keys and service passwords after any suspected exposure, including ones in old backups and exports. For a small business, that means connected-app tokens, API keys and shared logins.
Turn on sign-in alerts. Chainalysis told its users to watch for password-reset emails and sign-in prompts they didn't expect. Most CRM and email services can alert you to a new sign-in.
For your IT provider: per BleepingComputer, Grinich said WorkOS saw the user agents artex-enrich/1.0 and norma/0.4 in the attacker's web requests. CrowdStrike's hunting guidance lists command lines containing "artx" or "artex," artx-prefixed cloud roles, and the C:\ProgramData\artx5v\ folder. It also recommends requiring IMDSv2, enforcing SAML signature checks and rotating signing keys found in old directory exports.
What should you do this week?
Pick the one app that holds the most customer information, usually your CRM or email. Turn on two-step sign-in for every user, delete any account you don't recognize, and switch on new sign-in alerts. Then tell your staff: if a message about a real customer or job asks for money or a password, call back on a number you already have.
For more, see our guides to stopping fake payment requests, what to do when a vendor exposes customer data, and giving an AI assistant its own account instead of your login.
FAQ
Was my business's data in the WorkOS or Chainalysis incidents? Only if you are one of their customers. WorkOS says it will contact affected customers directly, and Chainalysis says it has notified all of its customers.
Is two-step sign-in by text message good enough? It beats a password alone, but CISA calls text and email codes the weakest kind. Use an authenticator app or a security key wherever the service offers one.
What is password spraying? It's trying a few common passwords against many accounts, slowly enough to avoid lockouts. Unique passwords and two-step sign-in blunt it.
Do I need AI security tools to stop AI attacks? Not to start. Our reading: the defenses Google lists, such as least privilege, short-lived credentials and separating systems, are long-standing basics that cover the weak spots these tools look for first.
Disclosure: No company or organization named in this article paid for it, reviewed it or saw it before publication. Cybercave Technologies builds, hosts and manages websites for small businesses and offers AI automation services. Incident details are as published by the companies and researchers named and were checked on Nov. 3, 2026. Not sponsored.
Sources
- PRIMARY: WorkOS, "October 30th, 2026 Data Security Incident," Nov. 2, 2026
- PRIMARY: Chainalysis, "Chainalysis Security Incident Information," Oct. 30, 2026, updated Nov. 2, 2026
- PRIMARY: Google Threat Intelligence Group and Mandiant, "ARTEX and Friends: Enter Agentic Offensive Frameworks," Oct. 30, 2026
- PRIMARY: CrowdStrike, "VAPOR PANDA Uses AI-Driven ARTEX to Grow Global Espionage Campaign," Oct. 29, 2026
- PRIMARY: Wiz, "ARTEX in the Wild: Storm-0940 Expands Targeting of IT and Software Vendors," Oct. 22, 2026
- PRIMARY: CISA, "Require Multifactor Authentication," accessed Nov. 3, 2026
- BleepingComputer (Bill Toulas), "WorkOS says attacker used ARTEX AI to breach internal sales app," Nov. 3, 2026
- The Hacker News, "Chainalysis Confirms Intrusion Into Internal Tools and Exposure of Customer Identity Data," Oct. 31, 2026