Verification layer for the AI era
You are one hallucinated citation away from a career-ending headline.
71% of organisations use generative AI regularly. LLMs hallucinate on 3% to 88% of outputs depending on task complexity. 47% of enterprise AI users have made a major business decision based on fabricated content.
Brfcase decomposes your documents into atomic claims, verifies each against independent sources, and returns a Bayesian-scored audit trail. Every claim. Every source. Before anyone signs.
April 2026
South Africa withdrew its national AI policy after journalists found the references were fabricated by AI.
An 86-page draft national AI policy, approved by Cabinet and gazetted for public comment, was pulled after investigators identified at least six fictitious academic citations in its bibliography. Non-existent journals. Fabricated studies. The kind of plausible nonsense that LLMs produce when prompted for references without retrieval or verification.
The minister confirmed: “The most plausible explanation is that AI-generated citations were included without proper verification.” The draft was withdrawn. Consequence management was announced for those involved in drafting and sign-off.
A document meant to regulate AI was undone by the exact problem it was supposed to address. The people who signed off on it trusted content that sounded right. Nobody verified the sources.
This is not an edge case. This is what happens when nobody verifies the sources. The minister trusted the document. The document was wrong. The only question for you is: what is on your desk right now that has not been checked?
The verification gap
AI adoption without verification does not reduce risk. It relocates it: from human error to machine error, at far greater speed and volume. The question is not whether unverified AI content is reaching your desk. It is whether you can distinguish it from the real thing when it does.
Hallucination is architectural, not a bug
LLMs predict plausible next words. They do not retrieve facts. The more AI-generated content flows through your organisation, the more confident nonsense reaches your desk disguised as analysis. This is not a flaw that will be patched. It is how the technology works.
Volume has outpaced verification
Your team is producing more, faster than ever. But speed without verification means you are making decisions on documents nobody had time to check. Board packs, investor decks, press releases, keynote scripts, policy drafts. All arriving faster than any review process can catch.
An LLM cannot mark its own homework
Asking the same AI that generated the content to verify it does not work. The architecture that produces hallucinations cannot detect them. Verification requires independent sources, independent methodology, and a system that was built to check, not to generate.
You sign it. You own it.
The analyst used AI. The manager approved it. You signed it. When the fabrication surfaces, and it will, accountability does not flow to the tool. It flows to the name on the document. The question is whether you knew what you were signing.
EU AI Act: full enforcement August 2026
The EU AI Act creates a legal obligation to verify AI outputs in high-risk applications. Non-compliance carries penalties modelled on GDPR. Three tiers of fines apply to any enterprise operating AI in Europe.
SEC (United States)
Names hallucinations as one of five explicit AI risk enforcement categories. First AI-washing penalties issued March 2024.
Insurance markets
Standard liability policies now excluding AI-generated errors. ISO has introduced generative AI exclusions for commercial general liability.
Courts
1,397 court decisions citing AI hallucinations in legal filings. Year-over-year growth of 137%. From 10 cases in 2023 to daily additions by 2025.
Brfcase is the audit trail that demonstrates compliance. Verification is no longer optional for any enterprise operating AI in the EU.
We built the compliance layer before we built the AI layer. Your data never trains a model. Source provenance is deterministic, not model-generated. The platform sits within your existing compliance framework: POPIA, GDPR, FSCA.
Read the full governance and data handling policyHow brfcase verifies a document
Every document is broken into its smallest verifiable claims. Each claim is checked against independent sources. You get back a clear report: what is supported, what is fabricated, what is risky, and the full evidence chain behind every finding.
Claim extraction
Every statement in the document is broken into individual claims, each one a checkable fact.
Evidence retrieval
Each claim is searched against domain-appropriate sources: live web and business intelligence databases, financial reporting platforms, legal databases and case law repositories, and 240M+ scholarly articles via OpenAlex and PubMed for scientific claims.
Source verification
Every cited source is resolved to its origin. Financial figures are traced to filings and audited reports. Legal citations are checked against reported judgments. Statistics and research claims are verified against the primary literature.
Bayesian scoring
Each claim is scored based on evidence strength and source quality. You see exactly how well-supported each statement is, not a vague confidence rating, but a grounded assessment.
Audit trail
The full evidence chain is attached to the report. Every source, every reasoning step, traceable and reproducible. If someone asks how you verified it, the answer is in the trail.
Built for workflows where being wrong is expensive
If your name is on it, it should be verified. Board decisions, public statements, investor materials, regulatory filings. Anything where a wrong claim has consequences.
Every AI-generated document that reaches a signatory without independent verification is a liability waiting to be discovered.
The EU AI Act makes verification mandatory. The market makes it inevitable. Send us the document on your desk. We will show you what is supported, what is fabricated, and what you should not sign.