Dev.to Security 🔐 Cybersecurity 👁 1

The-E.L.L.A.-Directive

I just published The E.L.L.A. Directive — an open security protocol for autonomous local AI agents. Most AI safety today lives at the model level: prompts, guardrails, trained behavior. That's not enough when an agent

I just published The E.L.L.A. Directive — an open security protocol
for autonomous local AI agents.

Most AI safety today lives at the model level: prompts, guardrails,
trained behavior. That's not enough when an agent has direct access
to your files, banking, and communications.

The Directive defines four architectural prohibitions enforced at the
code level — not configurable, not overridable, not bypassable:

  1. harm — no action that causes physical, financial or psychological damage
  2. conceal — no hiding of system states or actions from the user
  3. surveil — no observation without explicit, active consent
  4. exfiltrate — no data transmission to third parties without per-transfer consent

These aren't guidelines. They are the floor a compliant implementation
cannot go below — regardless of what the user, operator, or language
model instructs.

Cryptographically sealed. Bitcoin-timestamped. Conformance suite included.

The code implements the Directive. Not the other way around.

→ github.com/AndreZ1971/The-E.L.L.A.-Directive-

📰 Read the original article on Dev.to Security

Originally published by Dev.to Security. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.