Summary: In May 2026, CB Financial Services filed the first SEC Form 8-K under Item 1.05 caused by an employee's use of an unauthorized AI tool rather than an outside attack. Most fund managers will never file an 8-K, but the same facts at an adviser would trigger Reg S-P notification, exam questions and LP diligence follow-ups. The fix starts with knowing which AI tools are in use, using data your firm already collects.
What happened
According to a Wilson Sonsini analysis of the filing, the timeline was short:
- May 5, 2026: Community Bank, a Pennsylvania regional bank owned by CB Financial Services, learned that an unauthorized AI application had been used to process non-public customer information.
- May 7: The parent company determined the incident was material because of the volume and sensitivity of the data.
- May 11: The 8-K was filed. The exposed data included names, Social Security numbers and dates of birth.
There was no intrusion, no ransomware and no outage. Wilson Sonsini notes the bank called the incident material even though it did not disrupt operations and was not expected to affect financial results. The cause was an employee using an AI tool nobody had approved, most likely to save time.
Plaintiffs' firms announced investigations within days. Wilson Sonsini also flags a newer litigation theory: whether the firm had reasonable AI policies, whether it enforced them, and whether it had technical controls to back them up.
Why this matters to a hedge fund or PE firm
Most private fund managers are not public companies, so Item 1.05 does not apply to them. The underlying exposure does.
Reg S-P. The amended rule now applies to all registered advisers. It requires an incident response program and notice to affected individuals when sensitive customer information is accessed or used without authorization. An analyst pasting LP subscription documents into a personal AI account is that scenario.
Exams. The SEC Division of Examinations' FY2026 priorities say examiners will assess whether firms have adequate policies and procedures to monitor and supervise their use of AI. "We have a policy" will be followed by "show us what's in use."
LP diligence. Allocators read the same headlines. Expect the AI section of your next DDQ to ask how you would know if an employee used an unapproved tool with investor data.
How you would have seen it coming
Wilson Sonsini's first recommendation is to know where AI lives in the organization. In practice, most firms already hold the data needed to answer that question. It is spread across systems that were never asked to answer it.
| Signal you already have | What it shows about AI use |
|---|---|
| Web gateway or firewall logs | Which AI domains people visit, how often, from which devices |
| Identity provider / SSO | Which AI apps people sign into with corporate credentials, including OAuth grants to third-party tools |
| Endpoint management | Desktop AI apps and browser extensions installed on firm devices |
| Email and data protection | Files or text flowing to AI services |
| Expense and card data | Personal or departmental AI subscriptions paid for outside procurement |
None of these sources answers the question alone. A gateway log shows a domain, not whether the tool was approved or what data went into it. An SSO log shows a sign-in, not the purpose. The useful view comes from joining them: this person, in this department, used this tool, which is or is not in your approved catalog, under these terms.
What to do this quarter
- Build the inventory from signals, not surveys. An annual "which AI tools do you use?" survey only captures the tools people remember and are comfortable declaring. Start from the logs above.
- Compare it to what you approved. Every tool in use falls into one of three buckets: approved, under review, or unknown. The unknown bucket is your shadow AI.
- Give people a fast way to ask. Most shadow AI is a productivity decision made because the approval path takes weeks. If a request takes hours instead, fewer people go around it.
- Classify the data. Decide which data types (LP records, personal information, MNPI, deal terms) are off-limits for any tool that hasn't been approved for them, and write that into the approval.
- Update the incident response plan. Add unauthorized AI use as a scenario, with an owner, an escalation path and the Reg S-P notification steps.
- Keep the record. Every discovery, request, approval and exception should leave a trail you can hand to an examiner or an allocator.
Where Clarier fits
Clarier reads from the systems listed above (identity, endpoint, network, data protection and your model providers) to build a continuous inventory of AI in use across the firm, then matches each tool to your approved catalog. Unapproved tools surface as discoveries that route into the same request and approval workflow as everything else, so the record of what you found and what you decided sits in one place.
If you want to know where your program stands today, take the AI program maturity assessment.
