Rights and contracts
Contact center AI and third-party eavesdropping claims: the capability test
By SourceX Editorial · Reviewed by Noah Loul ·
Short answer
The capability test asks whether an AI vendor processing a business's calls or chats is able to use that content for its own purposes, such as training its models. If so, some courts treat it as a third-party eavesdropper under California's Invasion of Privacy Act. The narrower extension test asks only what the vendor actually does with the content.
Key takeaways
- Eavesdropping claims against contact center AI usually target the vendor as an unannounced third party on the call or chat.
- The extension test asks what the vendor actually does with the content; the capability test asks what it is able to do.
- Many rulings come at the pleading stage, so they decide whether a case continues, not whether anyone broke the law.
- Vendor terms that reserve rights to use conversations for model improvement are central evidence under the capability approach, period by period.
- Licensing historical recordings is not real-time interception, but a questioned capture travels with the records, so scope by period and prefer transcripts and agent notes.
What is the capability test in contact center AI cases?#
The capability test is one way courts decide whether a software vendor that receives a business's calls or chats is a third party under the California Invasion of Privacy Act. It asks whether the vendor is capable of using the conversation for its own purposes, such as improving models it sells to other customers, rather than whether it actually did.
These claims usually arrive as proposed class actions naming the vendor, the business or both. The business is a party to the conversation and cannot eavesdrop on itself, so the dispute centers on the vendor providing transcription, agent assist, quality scoring or a chatbot. California requires the consent of all parties for many recordings, which is stricter than the federal one-party baseline, and that is why so much of the litigation runs through California courts.
For a software company with a support line, the question is no longer confined to old-style call recording. Any tool that receives a conversation in real time and processes it on its own systems can be drawn in.
Capability test versus extension test#
Courts deciding whether a vendor is a third party have generally applied one of two approaches, and federal courts in California have not been uniform. The outcome in a given case can turn on which approach the court applies and how the complaint describes the vendor's technology and terms.
Procedural posture matters when reading the headlines. Many of these decisions come on a motion to dismiss, where the court asks only whether the complaint plausibly states a claim. A ruling that lets a claim proceed under the capability approach is not a finding that the vendor eavesdropped, but it does bring discovery, cost and settlement pressure.
The difference between the tests matters because many AI vendors never do anything with customer conversations beyond serving the business, while their contracts or product terms allow more.
| Question | Extension test | Capability test |
|---|---|---|
| Core question | Is the vendor acting as an extension of the business, like a tape recorder? | Is the vendor capable of using the content for its own purposes? |
| What makes the vendor a third party | Actually using the content for itself, such as to train its own models or sell insights | Having the ability to use it, whether or not it does |
| Role of the vendor's terms | Relevant to show what the vendor in fact does | Central evidence of what the vendor is able to do |
| Facts that help the defense | Content used only to serve the business and kept separate | Terms that bar any vendor use and technical separation |
| Facts that help the plaintiff | Evidence the vendor used content to improve shared models | Terms permitting model improvement or product development |
Why vendor terms sit at the center of these cases#
Vendor terms sit at the center because they are the easiest evidence of capability. Many conversation intelligence and contact center platforms reserve the right to use customer content, often described as service data or usage data, to improve or train their models, sometimes across all customers.
Under the capability approach, that reservation can support an argument that the vendor is a third party even if it never exercised the right for a particular customer. Opt-outs, enterprise addenda that switch off model training and data processing terms that restrict the vendor to serving the business can all change the picture.
Disclosures matter too. A greeting that says calls may be recorded for quality and training purposes was usually written with agent training in mind. Whether it gives adequate notice of a separate vendor, or of use to train AI models, is a question counsel should not assume away.
What the cases mean for licensing recordings and transcripts#
Licensing historical recordings or transcripts to an AI developer is a different act from real-time interception, and wiretap theories generally focus on the moment a conversation is captured. The capability cases still matter to a licensing decision, for three reasons.
First, wiretap statutes, federal and state, can also reach the later use or disclosure of contents that were unlawfully intercepted, so licensing records from a questioned period extends that exposure. Second, the vendor that captured the calls may hold its own rights in the content, and its terms control whether and how the business can export it. Third, a buyer that reviews provenance will ask what callers were told and which vendor terms applied.
Raw audio adds a separate layer. Voices can identify people, and state biometric privacy laws such as Illinois's Biometric Information Privacy Act, which allows private suits for liquidated damages, can come into play where voice data is used to identify individuals.
| Record type | Added concern | Typical approach |
|---|---|---|
| Raw call audio | Voices can identify people and may raise biometric privacy questions in some states | Often excluded or limited to periods with clear disclosures |
| Call transcripts | Names, account details and addresses spoken aloud | De-identified text from periods with documented disclosures |
| Website chat and messaging logs | Similar eavesdropping theories have been raised for chat vendors | Same disclosure and vendor-terms review as calls |
| Agent-assist suggestions and quality scores | Vendor-generated content that may belong to the vendor | Check vendor terms before including |
| Ticket notes written by agents after the call | Company-authored records, not captured conversations | Usually the lowest-risk part of the archive |
Reading vendor terms through the capability lens#
Reading each vendor's terms through the capability lens shows which periods of an archive carry the most exposure. The same platform can sit on different sides of the line over time as terms, settings and addenda change, so record the effective date of every version.
The table shows how common wording reads under that lens, and the steps below turn it into a period map before anyone scopes a package.
- List every system that touched conversations: telephony, contact center platform, conversation intelligence tool and chatbot vendor.
- Pull each vendor's terms, addenda and settings history, with effective dates.
- Mark each period by capability signal: reserved rights, training switched off, or instructions-only processing.
- Overlay the greeting scripts, chat banners and privacy notice versions in force for each period.
- Scope only periods with a low capability signal and documented disclosures, and prefer de-identified transcripts and agent notes over audio.
| Wording in vendor terms | Capability signal | What to record |
|---|---|---|
| Vendor may use Customer Data to provide and improve the Services | Strong if improvement reaches models shared across customers | Dates the wording applied and any product-level carve-outs |
| Vendor may use aggregated or de-identified data for any purpose | Moderate; depends on whether conversation content feeds it | The vendor's written answer on what the derived data contains |
| Model training can be switched off in settings | Reduced from the date it was switched off | Admin log or screenshot showing the date |
| Vendor processes Customer Data only on Customer's documented instructions | Weak; looks like an extension of the business | Signed addendum and its effective date |
| Vendor retains recordings or transcripts after the contract ends | Shows the vendor holds content beyond serving you | Retention period and any deletion confirmation |
| Subprocessors listed on a web page, updated from time to time | Others may also have had access | Saved versions of the list for each period |
Illustrative: a restaurant payroll software company maps vendor periods#
Illustrative: a fictional payroll and scheduling software company for restaurant groups runs phone support through a cloud contact center with a conversation intelligence add-on, and website chat through a separate chatbot vendor. Agent case notes live in Salesforce Service Cloud. A buyer has asked about licensing support conversations.
The general counsel sends both vendors written questions and builds a timeline. In the earlier period, the conversation intelligence vendor's terms allowed it to improve its models with customer content and the greeting mentioned only quality and training. In the later period, the company had signed an enterprise addendum switching training off and had updated its greeting to name third-party technology. The chatbot vendor's terms reserved rights to use chats throughout.
The company excludes all audio, every chatbot transcript and the earlier period's call transcripts. It offers de-identified call transcripts from the later period plus agent notes from every period, each tagged with the terms and greeting in force. The CEO approves the scope with the exclusions listed.
How SourceX approaches recorded conversations#
For recorded conversations, SourceX starts with how the records were captured and which vendor terms applied at the time. That review belongs to Rights in the SourceX five-step transaction (Supply, Rights, Preparation, Approval, Delivery), and the supplier and its counsel decide which periods and record types are in scope. Nothing is shared during the initial fit check.
Transcripts and notes that proceed have personal and confidential details removed in Preparation. The period map, including the greeting, banner, notice and vendor terms in force, is noted in the privacy record of the SourceX Evidence Packet, so a buyer reviewing provenance can see it without asking for the recordings themselves.
Frequently asked questions
Does the capability test apply outside California?
The test developed mainly in cases under California law, but other states have their own wiretap statutes and courts there may face similar questions. Federal law and many states generally allow recording with one party's consent, while some states are stricter. Counsel should assess the states where callers were located.
Is the business or the vendor the defendant?
Often both. Plaintiffs may sue the vendor as the alleged eavesdropper and the business for allowing it, or for aiding it. Contracts between the business and the vendor usually address indemnities and cooperation, so review them alongside any decision about licensing the content.
Does switching off vendor model training fix past periods?
Generally it changes the picture only from the date it took effect. Conversations processed while the vendor's terms reserved training rights keep that history, which is why the period map records when each setting or addendum changed rather than treating the archive as one block.
Can a buyer of our recordings be drawn into these questions?
Possibly. A buyer using content from a questioned capture may face its own questions about using that content, so buyers often ask for provenance records and rights warranties. Scoping the package to well-documented periods protects both sides and makes those warranties easier to give.
Do agent notes carry the same risk as recordings?
Generally less. Notes written by agents after a call are the company's own records rather than captured conversations, so eavesdropping theories fit them poorly. They can still contain personal information, so they go through the same privacy preparation before licensing.
Sources
- The Illinois Biometric Information Privacy Act (740 ILCS 14/20) lets a prevailing party recover liquidated damages per negligent or intentional violation, or actual damages if greater. Source
Related resources
See if your company qualifies
A short company assessment. No data uploads are needed.