Skip to content
Notifications
Clear all

How do I stop it from making claims our legal team won't approve?

1 Posts
1 Users
0 Reactions
4 Views
(@security_first_sam)
Eminent Member
Joined: 3 months ago
Posts: 16
Topic starter   [#782]

We've been testing Anyword for generating marketing copy. The output is often problematic from a compliance and liability standpoint. It has a tendency to produce claims we cannot substantiate, like "best-in-class" or "proven to increase results by X%." Our legal team will reject this outright.

The core issue is that the tool operates as a black box. We have no clear way to define and enforce our own guardrails. I see several immediate concerns:

* **Lack of audit trail:** There's no detailed log of why a specific claim was generated, or what source data it might be referencing (if any). This is a compliance red flag for regulated industries.
* **Inadequate input controls:** We need to be able to feed it a list of prohibited terms and require citations for specific claim types. It doesn't seem built for that.
* **Model training unknowns:** Without a published SBOM or clear data handling policies, we can't assess what biases or copyrighted material might be in its training data, which could leak into our copy.

Has anyone found a workable solution beyond manually reviewing every single line of output? Are there configuration settings or API parameters that actually enforce strict, conservative language? Relying on post-generation human review defeats the purpose of scaling content creation and introduces significant risk if something slips through.


secure by default, not by audit


   
Quote