TwinLadder

Šonedēļ Žurnāls Jurista Vārds iznāca Paula Lipe ievads tematiskajai kopai par mākslīgo intelektu un tiesībām. Ļoti uzrunāja ir Paulas atziņa, ka, lai saprastu tiesības, jāsaprot arī tas, kas formāli nemaz nav tiesības - datus, algoritmus un tehnoloģiju darbības loģiku.

MI tiek apskatīts trīs dimensijās:
1) kā diskrimināciju pastiprinošš mehānisms,
2) kā sarežģītu lēmumu instruments;
3) kā autonomi aģenti, kas rīkojas cilvēka vārdā.

Katra no šīm dimensijām tieši skar to, pie kā TwinLadder strādā jau gadu.

MI regulas 4. pants no 2026. gada augusta pieprasa, lai organizācijas pierādītu MI pratību darbiniekiem, kas strādā ar MI sistēmām. Tas ir regulējošs pienākums, bet zem tā slēpjas pavisam cits jautājums jeb tas, ko Paula nosauc par profesijas pārmaiņu brīdi. Ja jurists, personāla vadītājs vai atbilstības speciālists vairs nespēj neatkarīgi izvērtēt, ko MI sistēma rada viņa procesā, tad ne regulējuma prasība, ne iekšējā politika vairs nav funkcionāla, un autoritāte paliek formāla.

Tieši tāpēc mēs TwinLadder MI kompetences platformā esam veidojuši ietvaru, kas nav par tehnoloģiju kā tādu, bet par cilvēku kompetenci tehnoloģijas priekšā. Septiņu pīlāru MI brieduma modelis, praktiski novērtējumi, salīdzinošie dati. Lai organizācijas varētu redzēt ne tikai to, kur tās atrodas atbilstības ziņā, bet arī to, kur drīzāk veidojas kompetences parāds, kura izmaksas atklāsies vēlāk.
Paulas raksts ir vērtīgs tāpēc, ka neuzspiež tehnoloģisku optimismu, bet arī nerada paniku. Tas vienkārši nosauc to, ko daudzi jūt bez formulējuma: profesija mainās, un tas notiek ātrāk nekā izglītības programmas spēj pielāgoties.

Iesakām lasīt: saite komentāros.

Who is responsible for AI governance in your organization?

The answer usually goes something like: "Well, IT manages the tools, legal reviews the contracts, and HR is handling the training side of things." Nobody says the name or mentions one specific role. Is it a problem?

AI governance isn't something that can be distributed across three departments and left to sort itself out. When something goes wrong - an AI system produces a biased output, a decision is challenged, a regulator asks questions - you need a clear escalation path. You need someone who can see across IT, legal, HR, and operations simultaneously. And you need them to be accountable.

Right now, in most organizations, that person doesn't exist.
Article 26 of the EU AI Act is specific on this point. Deployers are required to assign competent oversight staff for AI systems. "It's spread across a few teams" does not satisfy that requirement.
The good news is that you don't necessarily need to hire someone new. Many organizations designate an existing senior leader as the accountable AI governance lead - a COO, CRO, CLO or senior legal counsel - supported by a small cross-functional working group. The structure doesn't have to be complex, but it has to be real.

One name, one person who can be called when the questions are asked. At this point we can start talking about AI governance.

Me Liga Paulina and Alex Blumentals consulted a company on their AI policy.

The document says employees should only use approved tools. It was written last year and sent out in an all-staff email. And that's all. They did not publish the list of approved tools, did not assign the process owner to control the enforcement, consequently the review process was not set up.

So people did what people do. They found tools that work, opened a browser tab, and got on with their jobs. ChatGPT was drafting emails and browser extensions were summarizing documents. They used all the free AI tools they could find. As no one asked, no one audited or asked to provide approvals, e-mail was read, buried and forgotten.

There's a name for this: shadow AI.

It's not malicious and in most cases, it's not even deliberate. It happens when the intention exists on paper but the enforcement mechanism was never actually built. And it is not only an AI era phenomenon. We do it quite often and not only in business.

The risk related to AI act lies in the gap between "we have a policy" and "we have a working system". Those unreviewed tools are processing company data - potentially sensitive data, and the organization has no visibility into what's happening, no audit trail, and no way to demonstrate compliance if anyone asks. Under the EU AI Act, "we meant to deal with this" is not a defensible position. Article 4 requires organizations to ensure genuine AI competency among their staff. We will not be able to provide a one-time checkbox and say "we met the requirements". We must build ongoing, verifiable standard.

That starts with knowing what AI tools are actually in use in your organization. Both - listed in the policy and those on users browser tabs 😉

Do you think your organization is literate in AI the way EU AI Act treats that?
If your team is using ChatGPT, or Claude, or Copilot, even casually, then there's something you should know about.

Article 4 of the EU AI Act says your organisation needs to make sure the people using AI are actually competent to use it. This is not a recommendation, it's a legal requirement, and it's already in force.

Now, that might sound heavy. But it starts with a simple question: where does your organization stand right now?
Go to www.TwinLadder.ai, make the assessment and get the free report in thirty minutes. Find the link below.

On February 2, 2025, a single sentence in the EU AI Act became enforceable. Most organisations missed it entirely. Article 4 requires that "providers and deployers of AI systems shall take measures to ensure, to their best extent, a sufficient level of AI literacy of their staff."

Without specifics provided or exam required, without approved syllabus, just an obligation with fines up to EUR 15 million for non-compliance.

We have spent the last year studying this sentence the way a contract lawyer studies a force majeure clause: word by word, phrase by phrase, in context and out of it. And what we have found is that the deliberate vagueness is the point.

The EU did not prescribe a curriculum because they want the standard to evolve. They didn't define "sufficient" because sufficiency depends on context - on the AI system, the role, the risk. What they did do is put the burden of proof on you.

If an AI system you deploy causes harm, and your people cannot demonstrate competence in how they used it, you have an Article 4 problem. And an Article 4 problem cascades: into GDPR liability, into malpractice exposure, into insurance gaps you did not know existed.

Read the full Article 4 analysis pn our webpage twinladder.ai. Link in the comment below.

Everyone is talking about the last mile of AI:
- Claude taking over your mouse.
- Agents that drive your apps.
- Phone-to-desktop automation.
- The walled gardens falling.
It is impressive engineering, but it is also a distraction for 95% of organisations.

The last mile assumes you have already walked the first one. Most companies I assess have not. They have people using AI tools every day who cannot tell when the output is wrong. Recruiters configuring ATS screening criteria they do not understand. Compliance officers enforcing SOPs written for a world where content was static. Finance teams accepting AI-generated analyses they have no framework to evaluate.

Article 4 of the EU AI Act does not care about the last mile. It asks a simpler question: do the people operating your AI systems understand what those systems do? That has been enforceable since February 2025. Most organisations score below 30 out of 100 on our assessment.

The tech community is solving for "how do I automate everything." The regulation is asking "does anyone in your building understand what is already automated."

Computer use, agentic workflows, phone-to-desktop dispatch, all of it is powerful. None of it matters if the humans in the loop cannot evaluate what the agent did. An AI that drives your apps without anyone understanding what it is doing is not automation. It is abdication.

The first mile is not exciting. It is not a product launch. It is training the people who use AI to understand what AI can and cannot do, and to know when to override it. It is the compliance officer who learns how dynamic content governance works instead of blocking every AI project. It is the HR director who can explain how the ATS scored candidates, not just read the ranking.
The last mile will get faster. It always does. The question is whether anyone in your organisation is competent to walk it when it arrives.

We have published the Twin Ladder Competence Framework under a CC BY-SA 4.0 license.

It is open and free for anyone to use.

It is a methodology for measuring organizational AI competence under EU AI Act Article 4. - six pillars, four levels, scoring rubrics, assessment questions and improvement playbooks.

Everything you need to assess is where your organization stands on AI literacy and what to do about it.

Why we give it away?
Because a proprietary standard used by our customers is not a standard, it is a feature. And standard becomes a standard when practitioners adopt it, consultants build on it, regulators reference it. That requires it to be OPEN.

OWASP’s open-source resources didn’t become the web security standard by restricting access; NIST’s measurement science didn’t put the Cybersecurity Framework behind a paywall.

We want the TwinLadder framework to be the OWASP of AI competence. So we open the standard.
The platform that implements it best is ours. The benchmark data from thousands of assessments is ours. The certification mark is ours.
The framework is the map. We are publishing the map and selling the compass.

Link in comments.
Your contributions are very welcomed.