AI, Power, Politics and the Battle for Narrative Control
Written by Nicolas Genest, CEO and Founder of CodeBoxx
Whether we admit it or not, AI is already inside our political and regulatory nervous system.
Campaign teams use it to test messages. Lobbyists use it to flood consultations with polished submissions. Activists use it to generate arguments at scale. Grants are documented and proposed using generated content and so are the corresponding grant applications and then the same models sometimes review and qualify the applications. In our democratic societies, with the blessing of a vast majority of citizens who immediately embraced generative AI three years ago to augment themselves and their impact, policy discourse is now shaped, accelerated, and sometimes distorted by systems no one ever voted for.
We absolutely don’t have an “AI innovation” problem anymore. We have an intent and governance problem. The models are all powerful enough and new versions reinvent obsolete benchmarks every other week. The missing piece is a clear answer to a simple question: Who controls the stories these systems tell, and on whose behalf?
Large language models don’t wake up one day and decide to be political. They are prompted to be and, at the time of training, inherited the views we fed them with. The data may be factual; the world it reflects isn’t fair. Models don’t invent bias; they reflect inequality, amplify it, structure it, and deliver it at machine speed. Societal insights about wellbeing, labor, demographics, family or consumerism become perceptible faster to more people through generated content the same way the 2000’s Internet revolution enabled information to travel at the speed of thought.
That’s why the aspiration to a “neutral” AI is dangerous. The moment you declare a system neutral, you’ve used free will to deliberately hide the value choices inside it: whose history counted as ground truth, which voices were over-represented, which harms you decided were acceptable side effects. When companies boast they’ve “reduced bias by 30%”, the only serious response is: according to which worldview?
Since November 30th 2022, some citizens exposed to reliable generative AI models see a potential for those tools to make politics better. First through bridging politicians knowledge, awareness and competency gaps. Then by evolving the legislative mechanisms from reactive policy making based on events, to proactive policy making based on a reliable ability to see projected impacts of long-term trends. And ultimately, by enabling all levels of government to exercise better judgment in the enforcement and application processes of the law. That’s the utopian promise of AI. But so far, three years later, the reported visible results of Gen AI penetration impacted narrative control and saw attempts to influence or disrupt democratic systems.
From aspiring to fake neutrality to communicating explicit intent.
AI Done Right must start with honesty, not perfection. Beyond the traditional terms of use and privacy policies of software platforms, we now need to add a statement of purpose for AI-Powered systems. If you deploy a system that will moderate speech, evaluate situations, rank information, make recommendations, draft legal text or advise on benefits, you owe citizens a clear statement of purpose: what is the service tool optimized for, what are the pursued outcomes, what data it relies on, what trade-offs were made, and who is accountable when it fails.
Every serious AI deployment should seriously consider carrying an intent label the way medicine carries dosage and side-effects. Governance should force these declarations into the open instead of letting them hide in model weights and marketing decks. These terms should also be negotiable when impactful institutions and organizations plan to commit to the use of such AI-powered tools and platforms.
From black boxes to traceable judgment.
Bias or will never be “fixed” away, not at its root, not at time of training and not at time of querying either and it shouldn’t be. It should be auditable. That means explainability and instrumentation built in from the first line of requirements, not bolted on after the PR crisis. AI Ethics injected into governance and policy making would mean asking: “If this system does exactly what we told it to, at scale, who gets hurt first?”
In practice, that would look like:
- Logging the decisions AI systems make and the signals they used.
- Documenting training data sources and known gaps.
- Giving affected people a meaningful and expedited path to contest outcomes
You don’t need AI that pretends to be objective. You need AI whose reasoning and judgment can be interrogated.
From automation fetish to human accountability.
Too many boards still talk as if delegating to AI reduces their responsibility. It does the opposite. When you put a model in the loop on evaluating, funding, credit, hiring, sentencing, welfare, or information ranking, you are intensifying your duty of care, not outsourcing it.
The governance test is simple: in every AI-mediated decision chain, can a named human explain what the system is doing, why it was deployed there, and how to shut it down if it starts drifting? If the answer is no, you’re not innovating, you’re gambling and maybe the currency is other people’s lives.
Upcoming elections and public debate will be a real stress test. As generative systems can fabricate voices, faces, and outrage on demand, the question is no longer “Will there be AI-generated propaganda?” It’s: Will citizens be able to tell which narratives emerged from humans, and which were weaponised by machines? That demands provenance standards, disclosure norms, and regulators who understand that information integrity is now a core safety issue.
The good news: we don’t need another miracle version of a branded large language model to fix this. We need agency, courage and clarity. Clarity from builders willing to expose their assumptions. Courage from regulators to demand traceability and intent. Not just asking for pretty dashboards but look at them and act on their insights. AI-augmented politics will require increased maturity from all of us to stop treating AI as mystical authority and start treating it as what it really is: a powerful mirror of our flawed societies we are putting between ourselves and our institutions.
In the end, AI won’t decide what kind of societies we become. People will still remain at the helm of civilization. It will be for the better only if we stay at the wheel of the narratives these systems generate, and force every model that enters public life to answer one blunt question:
What are you here to do – and who taught you that was the right thing?