When Machines Start Buying the News
Some years ago, machines learned to read the news. Today, they are beginning to understand it, and, crucially, to act upon it.
To understand this shift, we must look beneath the surface, into the plumbing and piping of AI and data infrastructures. There, a quiet but decisive transition is unfolding. Until now, AI has mostly been a passive tool serving human workflows. Humans were the bottleneck. Decisions were slow, negotiated, and debated. Machines assisted, but did not decide.
That is changing. In news media, a new model is emerging: News as a Service. In this model, autonomous agents communicate directly with each other. Soon, they will negotiate prices, conclude licenses, assess verification levels, and even propose or execute decisions. At that point, the scale and speed of machine-to-machine (M2M) interaction will exceed human comprehension. The tempo and volume of these interactions will be fundamentally beyond human capabilities.
If we are not careful, this leads to an uncomfortable asymmetry. We risk moving from shapers to observers, displaced by systems that operate faster and more consistently than we ever could. This shift is already visible in how news agencies plan to distribute information. Instead of sending massive wire feeds, a firehose of domestic news, politics, sports, and celebrities, 95% of which is irrelevant to an individual journalist at any given time, agencies are moving toward APIs. The agency API talks directly to the client API. A system might ask: "What changed in the last 24 hours regarding electric vehicle charging infrastructure in Germany, including references to subsidies and their semantic derivatives?" Machines are talking to machines about facts.
The traditional wire is quietly dying. Agencies are no longer limited to their own content either. Platforms like Reuters Connect already function as hubs, offering machine-readable content from a rich mix of external sources: AP, AFP, DPA, and specialised providers, delivered as structured data blocks. For now, humans still negotiate contracts and make final decisions. But the next step is already on the roadmap.
Reuters, for example, plans to add support for the MCP protocol, enabling deeper automation, scalability, and new licensing models. MCP is like the USB port for AI: a universal way for machines to plug into each other.
But interoperability alone is insufficient. Machines also need a shared language, an Esperanto for systems that want to understand and exchange news. That language exists. NINJS (News in JSON), developed by the IPTC, structures content so articles, photos, and videos become exchangeable, machine-readable data blocks.
Yet even then, machines still do not know what they are allowed to do with that content. For that, a next layer is required: machine-readable rights and permissions. Using RSL-1 descriptions, a news object can specify whether it may be republished, analysed, or synthesised by AI, and under which conditions. Compliance is then handled directly through the API.
To make this legally and economically watertight, the transaction needs a form of digital notary. That, too, already exists: smart contracts on infrastructures like Ethereum. Payments can flow automatically through APIs like Stripe, which already support fully machine-based payment decisions.
Taken together, this signals the emergence of a live machine-to-machine economy. It is governed by rules that are pre-defined, automated, and enforced at speed. Machines calculate confidence scores, verify provenance, assess risk, manage timing, and execute decisions. They do not debate like humans do. They execute.
Humans see only the result of the handshake. To avoid defaulting to the path of least resistance, simply accepting whatever the system returns, we will need significant critical thinking.
There are flashing warning lights here. An information provider, whether a government, a company or a media organisation, that does not speak this machine language effectively ceases to exist. If you are not machine-readable, you are invisible.
Whoever speaks the language, owns the territory. Trust and verification become variables in code. Provenance becomes metadata. Human intuition is replaced by confidence scores and cryptographic signatures. Without these, information is reduced to noise. This is not optional. If we do not invest here, we will not matter, and we certainly will not be paid.
This raises a deeper question: how does one machine know another is trustworthy? How do systems distinguish legitimate actors from parasites or even well-funded disinformation operations that technically comply with all protocols? If humans are removed from the loop, and we largely did that a decade ago with social media, then we need a public trust registry for AI agents and M2M actors. A DNS-like 'Chamber of Commerce' for machines. Here too, progress is being made. A registry originating at MIT, NANDA, is gaining traction. It records who operates an API or agent, under which legal identity, and with which governance rules. It tracks which agents trust or block others, existing audits, and applicable sanctions. Initiatives like NANDA form the keystone of this digital chain. They aim to combine DNS (identity), ISO-like standards (compliance), reputation networks, and legal registries into a single lookup designed for machines. Federated, non-profit, and sector-agnostic.
But the biggest warning light comes even earlier. What gives information providers genuinely high confidence scores? What earns the digital equivalent of a rubber stamp for trust? What sustains a golden reputation and avoids blacklisting?
For news organisations, the answer starts with an exhaustive data foundation. Everything that is societally critical must be ingested in structured form: public data, legislation, public health, traffic, weather, cybersecurity, mobility, finance, energy, and more.
On top of that foundation, journalism must do its work. In this emerging model, news is no longer a consumer product. It becomes critical democratic infrastructure. Journalists and editors monitor systems rather than isolated incidents; they connect dots, reveal blind spots, and verify facts. That work requires going outside, speaking to people, maintaining diversity of perspective, and applying ethical judgment, back to basics, but with new tools. The value is no longer filling news sites. It is detecting, verifying, and qualifying facts in ways that make them usable and trustworthy for society.
This makes journalism a public good, like road infrastructure or flood defence systems. It is therefore logical that governments become institutional clients of independent journalism, especially where regulation such as NIS2 explicitly requires monitoring, analysis, and alerting. But the same applies to any organisation subject to network and information security obligations, supply-chain risk management, or compliance. This concerns hard news: economy, government, energy, policy, technology, finance, and geopolitics. Not entertainment, emotion, or reputation management, but factual insight into what affects critical systems. The role of the news organisation becomes that of a trusted backbone: a reliable API partner for professional and institutional clients navigating a world where trust and insight are increasingly machine-mediated.
Under these conditions, we must prepare for a service economy in which the first users of our systems are autonomous machines. This is ultimately a sovereignty question. Who defines the protocols? Who sets the standards for liability and provenance? Whoever does, exercises power. If we do not lead in defining transparency and trust, others, including malicious actors, will.
Cyber territories are being designed right now, quietly, in the background. This is a silent revolution: the displacement of human decision-making. What counts as information? The headline in a newspaper, or what the most successful bot service returns?
We focus intensely on what disinformation does to citizens through social media and AI hallucinations. But what about incomplete or polluted information reaching Security Operations Centres, regulatory decision-makers, and infrastructure professionals? If machines negotiate reality on our behalf, who decides the rules of this conversation, and who still has the power to change them?
We are no longer just building tools for ourselves. We are designing the foundational protocols of the cyber territories of the future. Make sure you still have a passport to enter.
Synthetic image/AI-generated
This blog is written by Patrick Lacroix in a personal capacity. AI tools are used for research, structuring, drafting and language support. All content is selected, verified, and edited by the author, who retains full editorial responsibility.