The OPSEC Podcast: Privacy & Security for Life in the Digital Age

COVERT Protocol Action #4: Harden your Communications and Services

The OPSEC Podcast: Privacy & Security for Life in the Digital Age

Show Notes

AI assistants have been positioned as productivity tools. Structurally, they are the most sophisticated data collection interfaces ever built — ones that convince users to voluntarily disclose their most sensitive information in the form of natural language conversation. This episode covers what ChatGPT, Gemini, and Copilot actually do with your data, the corporate exposure epidemic (77% of employees are transmitting sensitive data to AI tools), how AI has supercharged the data broker industry through psychographic inference, and the government access problem that no privacy setting can fully solve.

Key Stats

4% of AI prompts and 20% of file uploads contain sensitive information (Harmonic Security, 2025) 3 million sensitive records exposed per organization by GenAI tools in H1 2025 77% of enterprise employees leak sensitive data via AI tools 67% of AI tool interactions happen on personal accounts IT cannot monitor 802,000 files at risk per organization in Microsoft 365 environments (Copilot access surface) 16% of business-critical Microsoft 365 data is overshared Psychographic AI inference: 70%+ accuracy predicting political, religious, psychological traits from behavioral data OpenAI privacy audit score: 48/100 (Grade D), 2026

Settings to Change Right Now

ChatGPT: Settings → Data Controls → disable "Improve the model for everyone" ChatGPT: Settings → Personalization → Manage Memory → audit and delete stored facts ChatGPT: Use Temporary Chat for sensitive queries Gemini: myaccount.google.com → Data & Privacy → Gemini Apps Activity → turn off Copilot: Microsoft account privacy dashboard → review AI data settings

Key Takeaways

AI interfaces are designed with conversational warmth specifically because it increases disclosure — that design is a data collection strategy Deletion does not guarantee destruction: the May 2025 federal court order proved ChatGPT "deleted" conversations can be preserved under legal hold Default on every major AI platform is collection; opt-out requires navigating settings most users don't know exist The corporate exposure problem is structural — 67% of AI usage happens on personal accounts IT cannot see AI psychographic inference manufactures sensitive personal data from behavioral signals — no explicit disclosure required Privacy-respecting alternatives exist: Confer and Lumo for cloud AI, Ollama for local maximum-security work Three-tier framework: Confer or Lumo → Ollama for sensitive work → commercial platforms only when the trade-off is consciously accepted

Every prompt you type into ChatGPT, Gemini, or Copilot is a data transmission you didn't think of as one. The lawyer who pasted in the brief. The HR director who described the investigation. The founder who uploaded the cap table. None of them made a mistake by their own understanding — but they all made one.

Stop treating AI interfaces as private spaces. They are not. Use the tools that were actually built to be — and make that decision before the conversation you can't afford to have retained.

Your privacy and your security is your responsibility. Hosted on Acast. See acast.com/privacy for more information.

More from The OPSEC Podcast: Privacy & Security for Life in the Digital Age

Oct 19, 2026

In this episode, I sit down with a political insider who has spent over 20 years working elections at the state and national level, plus consulting with governments and NGOs on public policy. He's been inside the machine that feeds the data swamp, offering answers to the voter privacy failures and campaign vulnerabilities I can only gesture at from the outside.

Oct 5, 2026

# The Voter File: Voter Data Brokers, Political Profiling, and Protecting Your Privacy You registered to vote. In nearly every state, that created a public record that voter data brokers can buy, enrich, and resell. This episode follows your registration from the county clerk into a private industry that fuses it with everything else known about you, builds political profiles that score your beliefs and your money, and sells the result. Roughly two hundred million registered Americans are in these files, and none of them consented to any of it. This is not about parties or who you vote for. It is about your data. ## In This Episode - What your state voter file actually contains, and the one thing it never contains: who you voted for. - Why the public half of the record is already enough to find you, address you by name, and reach you at home. - How a public record becomes a product through four layers: acquisition, standardization, enrichment, and modeling. - Enrichment: how commercial consumer data gets bolted on, with an industry claim of up to 1,800 fields per person. - Modeling: how a stranger's algorithm assigns you scores and predicted positions, then sells the guess as a fact about you. - What actually gets sold: universes, contact lists, scores, survey panels, and match services. - Why this is not a one-side story, and why the newest buyer, government agencies purchasing through commercial brokers, changes the threat. - Why the file leaks at the vendor layer, not the state, including the contractor that left modeled profiles on roughly 198 million Americans open on the internet. - Why the "non-commercial use only" restriction is an attestation, not a control. ## Key Takeaways - You cannot fully opt out of the voter file. Anyone who says otherwise is selling something. But you can shrink what makes it dangerous. - The real threat is not that you are a target. It is that your enriched profile lands in an investigation or a database that has nothing to do with you. - The one genuine opt-out is an Address Confidentiality Program (Safe at Home), and most people are not eligible. Check anyway. - The part you can actually fight is the enrichment layer. Break the commercial data bolted onto your record and a rich dossier becomes a thin one. - Stop feeding the same address into everything else, so your registration becomes a dead end instead of a master key. ## Do This Week 1. Look up your state's Address Confidentiality Program through your Secretary of State and find out if you qualify. If you do, enroll. 2. Start or restart your data-broker removal to attack the enrichment layer. 3. Separate your registration address from your car, deliveries, accounts, and loyalty cards where you legally can. ## Resources - U.S. Election Assistance Commission, voter lists, confidentiality, and list maintenance: https://www.eac.gov/election-officials/voter-lists-registration-confidentiality-and-voter-list-maintenance - NNEDV Safety Net, Voting and Survivor Privacy: https://www.techsafety.org/voter-registration-privacy - Your state Secretary of State website: search "[your state] Address Confidentiality Program" or "Safe at Home." ## Connect Subscribe and follow wherever you listen, and find every link at OPSECPodcast.com. Sign up for the newsletter at OPSECPodcast.com for the latest tools and tradecraft. Remember: your privacy and your security is your responsibility.

Sep 21, 2026

Ghost Signal: Ghost Signal: AI Voice Cloning Threats and Protecting Your Family From Digital Fraud For your whole life, a familiar voice has been proof of identity. A few seconds of audio from a social media clip is now enough to clone the voice of someone you love, and many cloning tools ask for little more than a name and an email address. This episode breaks down how AI-powered social engineering attacks get built, the three plays aimed at everyday people, and the free, low-tech protocol that beats all of them. In This Episode - What actually changed: AI removed the skill barrier from impersonation, and why "look for the typos" is now dangerous advice. - The assembly line: reconnaissance, pretext, voice clone, stacked channels, and manufactured urgency. - The three plays that do most of the damage: the family emergency call, the fake bank fraud call that asks for your code, and the message that knows your life. - How deepfake video is used to override a target's own suspicion. - Why the FBI now tells families to set up a secret word. - Why your family's public posts work as source material, and the military discipline of emissions control. Key Takeaways - A familiar voice or face is no longer proof of identity. Trust needs a protocol. - Every one of these attacks depends on urgency and on keeping you inside the attacker's conversation. Step outside it and the attack falls apart. - An AI can clone a voice from a video, but it cannot clone a word that was never recorded anywhere. - No bank, agency, or support line will call you and ask you to read back a code. If someone does, it is an attack. - A hardware security key cannot be talked out of you over the phone. - The people most likely to get the call are your parents and grandparents. Brief them with a specific rule, not a general warning.

Aug 24, 2026

Summary: Every financial transaction creates a permanent record: at the bank, the card network, the payment app, the data broker who purchased it, and the government reporting infrastructure mandated by the Bank Secrecy Act. This episode maps the complete architecture of financial surveillance—from 78,000 government reports filed daily, to Mastercard's transaction data sales, to Plaid's middleware granting hundreds of apps access to your full banking history. --- What Your Financial Data Reveals: - Health conditions (pharmacy patterns, clinic visits) - Political activity (donations, event venues) - Religious practice (charitable giving, places of worship) - Relationship status (dining, travel, shared accounts) - Financial stress (overdrafts, payday lenders, credit utilization) - Daily schedule and location patterns - Household composition --- Key Takeaways: - The Bank Secrecy Act generates massive surveillance as a compliance byproduct—most unrelated to actual crime - Structuring laws mean using cash incorrectly can itself be a federal crime, regardless of intent - Card networks (Mastercard, Visa) sell transaction data commercially; "anonymization" claims are disputed - Plaid aggregates your banking history and shares it with every connected app—most users don't know what they've authorized - Your financial profile is more detailed than your own memory of spending, accessible to multiple commercial and government actors --- Key Actions: 1. Audit Plaid connections: Visit plaid.com/connections to see every app with bank access. Revoke anything unused or unrecognized. 2. Get virtual cards: Use Privacy.com to generate merchant-specific card numbers with spending limits. Free tier available. 3. Read your bank's privacy notice: Exercise your legal right to opt out of third-party data sharing. Banks must offer it; few customers do. 4. Use cash strategically: For medical, personal, and sensitive purchases where you don't want a commercial record. 5. Don't structure deposits: Breaking up cash deposits below $10,000 to avoid Currency Transaction Reports is a federal crime—regardless of whether any underlying crime exists. --- Resources: - FinCEN: Bank Secrecy Act Overview - FinCEN: Currency Transaction Report FAQ - Privacy.com: Virtual Cards - Plaid: Manage Your Connections - PIRG: Mastercard Data Sales Practices - Cato Institute: BSA Reporting Volume FY2025 --- Call To Action: This week, go to plaid.com/connections and revoke every app you don't actively use. It takes five minutes, and you'll almost certainly find connections you forgot you created. Then read your bank's annual privacy notice and opt out of third-party data sharing. It's required to be offered. Exercise it. Your financial record is a behavioral profile. The question is who has access to it. Head to OPSECPodcast.com for everything else. Your privacy and your security is your responsibility.

See all episodes →