AI Detection Is Becoming a Business Decision
AI-generated content is no longer limited to experiments by technology companies. It is appearing in marketing copy, recruitment applications, customer emails, schoolwork, legal documents and publishing. A recent WIRED report examined Pangram, a young AI-detection company whose software estimates how much AI may have contributed to a piece of writing.
For you as a Malaysian SME owner, the issue is not whether your team uses ChatGPT or another writing assistant. The real question is whether you can make fair, informed decisions when software claims that text was produced by AI. A detector score may look precise, but it is still an estimate—not proof of misconduct, dishonesty or poor-quality work.
What Happened
Pangram became widely known after its CEO publicly shared high AI-detection scores linked to several disputed literary works. The WIRED report said Pangram identified one manuscript as 78 percent AI-generated, while other works received scores of 60 percent, 97 percent or even 100 percent. These allegations attracted attention in publishing and raised questions about how much influence an automated score should have over a writer’s career.
The company was founded in 2023 as Checkfor.ai and later renamed Pangram, according to WIRED. Its technology uses a process described as “synthetic mirroring”: the system compares human writing patterns with text generated by language models. It also uses “hard negative mining”, which means studying examples that could produce false positives and adding them to training data.
Pangram is not the only company working in this field. The article mentioned other tools, including Originality.ai, GPTZero and Turnitin. Pangram has also expanded into education, legal work and recruitment, while a Substack integration was announced so readers could check possible AI involvement in published writing. At the same time, writers and publishing professionals have warned that detector results can create serious consequences when people treat them as final verdicts.
Why This Matters for Malaysian SMEs
Your business may already be receiving AI-assisted job applications, proposals, sales emails, product descriptions and social media content. A recruitment manager might paste a cover letter into a detector and reject a candidate after seeing a high score. A manager might question a staff member’s report because it sounds unusually polished. A client might accuse your agency of using AI without permission after scanning a deliverable.
These situations are especially relevant in Malaysia’s multilingual business environment. A piece of writing may move between Bahasa Malaysia, English, Mandarin or Tamil, sometimes with translation software involved. A detector trained mainly on particular languages, writing styles or datasets may behave differently across those contexts. A score that appears authoritative may not reflect the writer’s actual process, especially when the text has been edited, translated or written by someone using formal business language.
Consider a local recruitment agency hiring a customer-service executive. A candidate’s English cover letter may be grammatically strong because they used a writing assistant to correct spelling and structure. That does not automatically mean the candidate lacks communication ability. Similarly, a small online retailer may use AI to create an initial product description, then verify specifications, rewrite the tone and add local delivery information. The final content may be useful and accurate even if software helped with the first draft.
Use an AI detector as a conversation starter or risk signal, not as the sole evidence for a disciplinary, hiring or client-facing decision.
Practical guardrails for your business
| Business situation | Safer approach |
|---|---|
| Recruitment | Assess skills through interviews, work samples and reference checks instead of relying on a detector score. |
| Marketing content | Review accuracy, originality, brand voice and disclosure requirements before publishing. |
| Internal reports | Ask for source links, calculations, drafts or explanations where verification is important. |
| Client deliverables | Agree in writing on acceptable AI use, human review and confidentiality expectations. |
| Disciplinary action | Never treat an automated score as the only basis for punishment or termination. |
How You Should Use AI Detection Tools
Start by defining what your company is trying to protect. Is the concern plagiarism, confidential information, inaccurate claims, impersonation, low-effort work or a breach of client instructions? AI detection may be relevant to one of these concerns, but it will not solve all of them.
For example, if your main concern is copied content, use plagiarism checks and compare source material. If the concern is inaccurate financial or product information, use a human approval checklist. If the concern is confidential customer data being entered into an external system, create a clear data-handling policy. A detector score may not answer any of these questions.
You should also document acceptable AI use. A simple internal policy can state that employees may use approved tools for brainstorming, translation, grammar correction and first drafts, provided they do not upload confidential information and remain responsible for the final result. For regulated, sensitive or client-owned work, require manager approval and human verification.
Ask staff to keep useful evidence of their work process when the assignment is important. This could include notes, research links, draft versions, source documents and review comments. Such evidence is generally more useful than a percentage score because it shows how the work was created and checked.
The Bigger Picture
The rise of AI detectors shows that businesses are entering a period where authorship, originality and transparency matter more—but simple labels will not be enough. A detector can identify patterns that resemble machine-generated text, yet resemblance is not the same as certainty. Writing style varies by industry, education, language, editing history and the use of translation tools.
The WIRED report also highlighted the reputational risk created when detector results are shared publicly before the facts are established. For an SME, the same principle applies internally. Accusing an employee, applicant or supplier based on an unverified score can damage trust and expose your business to complaints or disputes.
There is also a commercial question: what do your customers actually value? They may care less about whether an AI tool helped produce a draft and more about whether the final output is accurate, useful, original, secure and aligned with the agreed brief. If your business promises human expertise, you should be transparent about where automation is used. If AI assistance is permitted, focus on quality controls rather than pretending the tools do not exist.
Your next step is straightforward. Review where AI is already being used in your workflow, classify tasks by risk, set disclosure rules and train managers not to overinterpret detector scores. Treat detection technology as one input among several. The strongest protection for your business is not a percentage displayed on a screen; it is a clear process that combines responsible automation, human judgement and evidence.
Ready to Streamline Your Operations?
Technology moves fast. Your operations should keep up. AutoRunBiz builds AI systems that run your daily workflows — from WhatsApp order capture to accounting. Book a free 15-min ops audit →