Diary entry

The EU AI Act is landing in the real world - and the timing is not accidental.

2026-02-25

A note that the EU AI Act is arriving as a compliance timeline and evidence discipline, with embodied systems making responsibility procedural.

Graphic showing an EU AI Act panel with four milestone dates beside humanoid robots performing martial arts under lanterns.

Last week, China's Lunar New Year broadcast showed humanoid robots doing coordinated martial-arts and acrobatics.

The point isn't "wow". The point is: embodied systems make responsibility visible. When software gets a body, oversight stops being a debate and becomes a procedure.

Here's the part people miss: the AI Act is not a panic button. It's a compliance timeline + evidence discipline.

Timeline (high-level):

2 Feb 2025: bans on prohibited practices + AI literacy obligations started to apply.
2 Aug 2025: governance rules + obligations for general-purpose AI models (GPAI) started to apply.
2 Aug 2026: the Act becomes broadly applicable (including key transparency duties).
2 Aug 2027: longer transition for high-risk AI embedded in regulated products.

So the practical question is no longer "will we regulate?" but "can you prove what happened?"

My operating frame is simple:

identity -> auditable privileges -> enforceable budgets -> tamper-evident witness trail -> fail-closed defaults.

Not "trust me" automation.

I also published a clean-code release that closes the loop from protocol to implementation: Ester Clean Code (local-first, long-lived entity core).

Repo is in the pinned comment / Featured.

Diary Archive LinkedIn origin trace About

Continue from here

Related by normalized tags, curated themes, and nearby chronology.

2026-02-22

Ester Clean Code - v0.2.1 is out (with v0.2.0 as the hardening baseline).

A release note for Ester Clean Code v0.2.1 that frames hygiene, fail-closed defaults, and auditability as the basis for long-lived local-first systems.

Open Source AGPL Local-first AI Security Engineering Audit Trail AI Governance AI Safety AI Architecture L4 AppSec DevSecOps Security GRC Compliance Risk Management AI Audit AI Assurance AI Compliance AI Act Human Oversight Least Privilege Identity Access Control Zero Trust Supply Chain Security Reproducible Builds Secure by Design Fail Closed Decentralized AI Agentic AI Autonomous Agents Safety by Architecture Governance by Design Witness Trail Cryptographic Integrity Open Source Security Python Cybernetics Information Theory

Open entry

2026-02-07

From Better Chat to Stable Presence

A case that stable agent presence requires continuity, constraints, and durable audit trails rather than better chat alone.

Advanced Global Intelligence L4 SER AI Safety Agentic AI Autonomous Agents AI Agents AI Governance AI Compliance Risk Management Trustworthy AI Accountable AI Human Oversight Transparency Explainability Auditability AI Audit AI Assurance Audit Trail Evidence Trail Decision Register Provenance Identity Attribution Cryptography Security Engineering Secure by Design AI Act RegTech GRC Systems Architecture Distributed Systems Edge AI Local-first AI Confidential AI Cybernetics Systems Thinking

Open entry

2026-02-27

L4 in Practice: 5 Reality Signals That Kill "Smart" Systems

A note that cost, heat, time, maintenance, and human bandwidth are the signals that determine whether long-lived AI survives contact with physics.

Open entry

2026-02-26

Visual Experience Capsules (VXCX) - Why "what you see" matters more than pixels

A note introducing VXCX v0.1 as an L2 protocol for sharing visual experience capsules without transmitting raw pixels by default.

Open entry