Managing AI Slop: Why Due Diligence Is Essential for Business Leaders
Generative AI is now embedded across finance, legal, HR, operations, and comms. It drafts, summarizes, and produces analysis at speed. That scale comes with a quiet threat: AI slop-confident, fluent, and wrong content that slips into decisions, filings, and client work.
Recent incidents tied to AI-assisted reports in Australia and Canada showed how plausible-looking errors can pass review and end up in official submissions. The lesson is simple: without due diligence, AI becomes a liability. With it, AI becomes usable.
What "AI slop" is-and why it's risky
- Fabricated facts, citations, or legal references
- Misattributed sources or invented quotations
- Logical gaps hidden by smooth language
- Outdated or context-mismatched information
Large language models don't "know" facts. They predict likely text. Ask for citations, regulations, or research, and they can produce references that look correct but don't exist. That's survivable in a brainstorm. It's dangerous in a compliance memo, tax position, ESG disclosure, or government submission.
What recent failures make clear
In Australia, a consultancy report submitted to government included fabricated references and an invented court quote. In Canada, a publicly funded health workforce report cited studies that didn't exist. Both were challenged after submission.
The common breakdowns:
- AI-generated content made it into final deliverables
- Source checks were skipped or superficial
- Review focused on narrative flow, not factual integrity
- AI use wasn't clearly disclosed or governed
Why your standard QA won't catch it
Traditional review hunts for math mistakes, style issues, and policy compliance. It does not assume reality can be fabricated. A citation that "looks right" often sails through unless reviewers are told-and resourced-to verify every source. AI slop is a new failure mode: content that reads well and misleads.
Treat AI due diligence as enterprise risk
AI misuse and hallucinations belong in the same risk bucket as financial misstatements, regulatory breaches, cybersecurity incidents, and data privacy failures. That means moving from ad-hoc controls to formal oversight.
- Board and audit committee oversight of AI risk
- Internal controls that separate drafts from authoritative outputs
- Clear accountability for AI-assisted work products
- Vendor, partner, and professional liability alignment
- Client and stakeholder disclosure policies
If you need a reference framework for risk thinking, review the NIST AI Risk Management Framework.
Five principles to manage AI slop
1) Purpose-bound AI usage
- Define what AI can and cannot do: drafting assistance and summarization are allowed; fact generation, legal interpretation, or "original research" are not without expert review.
- Draw hard lines between assistive drafting, authoritative analysis, and mandatory human verification.
2) Human verification-line by line
- Any AI output with facts, figures, citations, legal or regulatory references must be checked by a qualified professional.
- Verification is source validation, not editing. If a source is cited, someone confirms it exists, says what the text claims, and is current.
3) Traceability and auditability
- Document which tool was used, for what task, by whom, and how it was verified.
- Keep version history and review notes. This protects the organization and the individual.
4) Transparency with clients and stakeholders
- Disclose AI use where it could affect reliance, interpretation, or liability.
- Bake disclosure into engagement letters and delivery templates.
5) Training and cultural alignment
- Teach teams how AI fails, not just how to prompt it. AI literacy is now a compliance skill.
- Reward accuracy and verification, not just speed and volume.
Industry checklists managers can use
Professional services (tax, consulting, governance, HR)
- Ban AI from issuing final tax opinions, legal interpretations, or compliance advice without senior sign-off.
- Independently verify all citations, statutes, case law, and regulations surfaced by AI.
- Record where AI assisted vs. where professional judgment was applied.
- Disclose AI use in client deliverables where relevant.
- Update professional indemnity to reflect AI-related risks.
- Include AI misuse in internal risk and audit reviews.
- Train staff on hallucination risks in regulatory content.
- Red flags: AI-generated references to laws, cases, or guidance; AI-assisted benchmarking or policy comparisons without source checks.
Technology companies
- Separate AI-generated docs from authoritative technical specs.
- Validate AI outputs used in filings, security docs, or compliance claims.
- Test AI outputs like you test software-repeatable checks and sign-offs.
- Add review gates for AI-generated customer-facing content.
- Keep marketing and investor materials free of unverified AI analysis.
- Disclose model limits and data constraints where required.
- Align AI governance with privacy and security controls.
- Red flags: AI-written safety or compliance documentation; AI-generated claims about regulatory status.
Research organizations and think tanks
- Forbid AI from generating original citations or academic references.
- Manually cross-check every source in AI-assisted drafts.
- Label AI-assisted sections in internal workflows.
- Hold firm authorship and accountability standards.
- Train researchers to spot fabricated studies and data.
- Protect reputations by preventing false attribution.
- Use publication review committees for AI-assisted outputs.
- Red flags: AI-written literature reviews; comparative studies without verified datasets.
Questions every board and ELT should ask
- Where is AI used across our business today?
- Do our controls separate drafts from authoritative outputs?
- Who is accountable if AI-generated errors reach clients or regulators?
- Are vendors and advisors meeting our due diligence standard?
- What's our training plan to build AI literacy for managers and reviewers?
Execution tips that work
- Set "AI use allowed/AI use restricted" labels by workflow. Make the default conservative for anything external-facing.
- Adopt a citation checklist: existence, accuracy, currency, context match. No check, no publish.
- Track AI-assisted content in your document metadata. Audit it monthly.
- Add consequence management: if it's signed, a human owns it.
From hype to discipline
AI isn't reckless-it's indifferent to truth. Your governance provides the spine. The high-profile failures weren't technology problems. They were verification problems.
Leaders who embed verification, transparency, and accountability will build trust with regulators, clients, and investors. Those who treat AI as a shortcut will pay for it in public.
If your team needs practical upskilling on safe, high-quality AI use by role, explore courses by job at Complete AI Training.
Your membership also unlocks:
