The Hidden Cost of Using AI at Work: Who Sees Your Prompts? | Ep 43

When you paste contracts, strategy docs, or internal PDFs into ChatGPT, it feels like you’re just talking to a smart assistant. In reality, you might be throwing your company’s trade secrets over the wall into someone else’s infrastructure.

In this episode of Inside the Silicon Mind, Firas talks with Confident Security founder and CEO Jonathan Mortensen about the hidden cost of using AI at work: what you send to AI tools is not end‑to‑end encrypted, may be retained, and in some cases can be used to train models. Jonathan has already sold one startup to a cybersecurity firm and another to Databricks; now he’s building “you bring the AI, we bring the privacy” – a private inference layer modelled on Apple’s cryptographic approach to Apple Intelligence.

Who this episode is for:

• This episode is for you if:

  • You’re a founder or executive using AI tools in workflows that touch sensitive docs, contracts, or IP.
  • You’re an engineer or security leader trying to understand the real risks behind AI prompts, model training, and prompt injection.
  • You’re a knowledge worker relying on ChatGPT at work and wondering what actually happens to the data you paste into it.

What you’ll learn in this episode:

• Why your AI prompts aren’t really private
  How today’s AI tools process prompts, why they’re not end‑to‑end encrypted, and why that matters for legal documents, strategy decks, and source code.
• How companies are training external models on their own trade secrets
  Realistic scenarios where sending internal PDFs and contracts to ChatGPT effectively hands those patterns to a third‑party model.
• What Apple’s Private Cloud Compute changes
  How Apple designed Apple Intelligence so that even Apple cannot see your data, and why publishing server code for public audit is such a big deal.
• How Confident Security works under the hood
  Why anonymisation (oblivious HTTP), remote attestation, and client‑side encryption together make it possible to run AI inference without exposing raw prompts.
• How a founder with two exits thinks about product and positioning
  Jonathan’s lessons on survivorship bias, “you’re in the business your customers think you are,” and choosing the difficulty of your business.
• The next wave of AI risk: deepfakes, prompt injection, and cognitive atrophy
  Why Jonathan won’t trust AI browsers to roam across tabs yet, and how over‑reliance on AI may blunt our intuition in the same way GPS changed navigation.