How to License AI Expert Advisors in 2026

Why AI EAs changed the piracy equation

Traditional Expert Advisors ship logic inside the compiled .ex4 or .ex5. Crackers patch the license check and redistribute the file. AI Expert Advisor licensing adds a second asset: trained model weights, ONNX exports, remote inference endpoints, or proprietary feature pipelines. A buyer who forwards only the binary might still leak the model, the API key, or the signal webhook.

In 2026, most "AI EAs" on MetaTrader combine MQL5 execution with one of these patterns:

1. On-chart ONNX inference — model file bundled or downloaded at activation
2. Remote signal API — EA calls your server each bar or tick for a score
3. Hybrid copier — Python service generates signals; MT5 EA executes

Each pattern needs a different EA license management strategy, but the business rule is the same: only paying customers may run your edge.

What to license beyond the key

A MetaTrader license key alone validates entitlement. For AI systems, also protect:

• Model artifacts — encrypt ONNX or weight files; bind decryption to activation
• Inference endpoints — rotate API tokens per customer; rate-limit by key
• Feature schemas — version the input vector so old cracks cannot replay stale requests
• Build binding — tie the EA binary to the model version you shipped

Without build binding, a cracker can attach your leaked model to an unprotected stub EA and bypass your storefront entirely.

Remote inference vs on-device models

Remote inference keeps IP off the client machine. The EA sends hashed market features; your server returns buy/sell/hold. Licensing lives on the API layer — revoke the key and calls fail within minutes.

Trade-offs:

• Latency and VPS uptime become part of your SLA
• Customers need stable outbound HTTPS from the terminal
• You must handle broker symbol suffix mismatches in feature extraction

On-device ONNX reduces latency and works offline after activation, but the model file is on disk. Combine file encryption, activation-bound decryption, and periodic online revalidation so expired keys stop loading weights.

Activation limits for AI bots

AI EAs often run on a dedicated VPS with tick-level inference. Standard retail limits still apply:

• 1–2 activations for solo traders (live + backup VPS)
• Higher seat pools for signal sellers running master + follower setups
• Stricter limits when the product includes API quota (calls per day)

Document whether one key covers one inference endpoint or one MT5 account. Mixing the two causes support churn when customers spin up second VPS instances.

Compliance and marketing claims

Regulators and prop firms increasingly scrutinize "AI" marketing. Your license terms should clarify:

• The EA is software tooling, not guaranteed returns
• Past backtests do not predict live performance
• Customer data sent for inference is handled per your privacy policy

Licensing metadata helps here — log activation region, broker, and product SKU so you can respond to abuse reports without guessing which build leaked.

Shipping a protected AI EA release

1. Train and version your model (v2026.08 in filename and build ID).
2. Protect the MT5 EA with MT5 EA protection and build binding.
3. Issue keys with activation limits and optional broker locks on pricing tiers.
4. If using remote inference, provision per-key API credentials at purchase.
5. Deliver via your site or marketplace escrow — never raw Telegram attachments.
6. Monitor activation spikes; revoke and publish a bound replacement if a model endpoint leaks.

Where this is heading

Expect tighter integration between licensing dashboards and inference quotas — keys that encode max daily API calls, model tier (lite vs pro), and automatic downgrade on expiry. Developers who treat AI weights like source code, and license keys like API credentials, will outlast one-file crack cycles.

If you are launching an AI-driven strategy this year, protect the full stack — binary, model, and server — before the first public sale.