The Sentence Nobody Wants to Engage With
Buried in the reporting on the Pentagon's May 1 AI deals is a single sentence that should change how you think about every American AI product you use. Anthropic was excluded from the Pentagon contracts because they refused to grant unrestricted access to Claude for "fully autonomous weapons and mass domestic surveillance."
Read it slowly. The Pentagon wanted commercial AI for mass domestic surveillance. Anthropic said no. OpenAI, Google, Microsoft, AWS, Nvidia, SpaceX, Reflection, and Oracle all said yes.
Every American who uses ChatGPT, Gemini, Microsoft Copilot, or any AI product from the eight companies that signed needs to understand what that means in practice.
What Mass Domestic Surveillance Actually Requires
Mass surveillance has historically been limited by a single bottleneck: humans cannot watch enough cameras, read enough emails, or analyze enough phone records to monitor millions of people at scale. Every previous surveillance regime — Stasi, KGB, Mukhabarat — required a labor force comparable in size to the population being surveilled.
AI eliminates that bottleneck. A single model can:
- Process every license plate read by every traffic camera in real time
- Analyze the content and metadata of every text message and call
- Cross-reference financial transactions with location data with social media activity
- Flag pattern-of-life anomalies that no human analyst would notice
- Build behavioral predictions about specific individuals
What used to require thousands of intelligence officers can now be done by one model running on a few thousand GPUs. The Pentagon just bought eight companies' worth of those models.
The Data Sources Already Exist
The intelligence community has been building data infrastructure for two decades. They already have:
- Cell tower location records from every major carrier (purchased through commercial data brokers)
- Social media activity (collected through partnerships and OSINT operations)
- Financial transaction data (Treasury Department FinCEN feeds)
- License plate readers (Vigilant Solutions, DRN, and others)
- Facial recognition databases (Clearview AI and similar)
- Domestic communications metadata (Section 215 collections, modified post-2015)
What was missing was integration. AI provides the integration. Eight companies just gave the Pentagon access to the AI.
What This Means for Your Personal Data
If you live in the United States and use the internet, your data is already collected by commercial data brokers and resold to government agencies. The FBI Director publicly admitted to this practice on the record. ICE has spent millions on Venntel location data. The Treasury and DOJ are major customers.
The integration of that data through AI changes the threat model. Previously, your data was somewhere in a database. Now it is being analyzed in real time, cross-referenced with other data, and used to build models of your behavior.
You cannot opt out of being in the database. You can only reduce the amount of data the database contains about you.
The Practical Defenses
The defenses that work in 2026 are operational rather than legal. The legal protections against mass surveillance have been weakening for two decades and the political will to strengthen them does not exist. Personal operational security is what is left.
1. Use a VPN to mask your IP and DNS lookups. Every site you visit, every API call your phone makes, every connection your laptop initiates — all of it is currently visible to your ISP, who sells the metadata to data brokers, who sell it to the federal government. A VPN encrypts that traffic and routes it through a server in a different jurisdiction, breaking the data correlation. NordVPN is what we recommend — Panama-based jurisdiction, audited no-logs policy, and Threat Protection that blocks known surveillance trackers automatically.
2. Audit your phone's location permissions. Every app with location access is a data leak. Settings → Privacy → Location → revoke access from anything that does not need it. Set the rest to "While Using App" instead of "Always."
3. Reset your advertising ID monthly. iOS: Settings → Privacy → Tracking. Android: Settings → Google → Ads. This breaks the cross-app tracking that data brokers rely on.
4. Use end-to-end encrypted messaging. Signal for personal communications. Avoid SMS for anything sensitive. Even WhatsApp metadata is collected, even though message contents are encrypted.
5. Use a privacy-focused email provider for sensitive communications. Proton Mail or Tutanota for anything you do not want indexed by Google or Microsoft.
6. Pay cash where possible. Cash transactions do not produce data trails. The data trails are what gets aggregated into the surveillance models.
Why a VPN Is the Foundation
Most other privacy improvements compound the value of a VPN. The VPN encrypts the network layer. Encrypted messaging encrypts the application layer. Permission audits reduce the data sources. But none of those work without the network-level encryption that a VPN provides.
Without a VPN, your ISP sees every domain you visit, every connection your phone makes, every video you stream. That data is collected, aggregated, and sold. Even with HTTPS encryption protecting the content of your traffic, the metadata about what you connect to is enough to build a detailed profile.
With a reputable VPN, your ISP sees encrypted traffic to a single VPN server. The metadata correlation breaks. Data brokers lose one of their primary data sources. The federal government loses one of its primary commercial data feeds.
That single change cuts more surveillance signal than any other personal action you can take. Set up NordVPN here — five minutes of configuration, then it runs automatically on every untrusted network you connect to.
The Honest Truth About Your AI Tools
If you use ChatGPT, Gemini, or Microsoft Copilot, you are now using AI tools built by companies that have explicitly agreed to support Pentagon mass surveillance applications. The same models that help you write emails are the foundation of intelligence analysis pipelines.
This does not necessarily mean your specific conversations are being surveilled. Commercial AI products and classified Pentagon AI deployments operate on different infrastructure with different access controls. But the same companies, the same engineers, the same training data, and the same underlying technology now serve both customers.
If you want frontier AI without that connection, Claude is the only frontier model from a company that has refused to support mass surveillance applications. That is a real distinction in 2026.
The Bigger Picture
Mass surveillance is not coming. It is here, in production, running on commercial AI infrastructure, paid for by Pentagon contracts, supported by eight of the largest tech companies in the world.
Personal operational security is what is left. A VPN is the foundation. Everything else builds on top.
