AI agents and cryptographic hashes: the 'agentic' newsroom is born

From the generation of content from scratch to the rewriting of texts that have already been elaborated: artificial intelligence has entered the world of communication 'in an avalanche', and the story, reported by The Verge, in which Google is said to be working to instruct its models to automatically modify the titles of articles directly in the search results page (in some cases altering their meaning) is symptomatic of the depth of the transformation we are experiencing. Journalistically speaking, it is a crucial step, because the headline is the first interpretative key of a news story and represents, in essence, a sort of implicit pact between the writer and the reader. That then the laws of the online audience (and thus of Web traffic produced by users) have somewhat undermined this convention we pretend to forget. The issue at the centre of attention is another: if automatic rewriting (of titles or even articles already published) becomes the practice, the risk is that of a progressive loss of control by editorial offices over their own editorial production, with potentially distorting effects on readers' perception of content. At the heart of the debate therefore remains the contrast between the quantity and quality of information, which is reflected on the one hand on the level of its veracity and reliability, and on the other hand sees the arrival in the field of a new generation of platforms based on AI agents capable not only of generating news, but also of certifying it.

The task of ProofPress, which presents itself to the Italia market as the first model of agent journalism capable of replicating and automating the entire workflow of a newsroom thanks to AI, is precisely that of producing certified information. If Verify, the algorithm developed in Silicon Valley in 2024 that validates every piece of news through Web3 'notarisation', has just landed in Italia thanks to Andrea Cinelli, already deux ex machina of the startup FoolFarm and the more recent Agentic Business Platform called Fasteer AI, the idea behind the platform starts from further back, and precisely from 2022, when a group of 15 insiders and nerds belonging to the startup and venture capital ecosystem gave birth (between Milan, Lisbon, Berlin and Tel Aviv) to IdeaSmart, a network of bulletin boards for internal use defined with originality (it was at the end of the pandemic period) as an 'information survival tool'. The following year, generative AI came into play and with it gradually took shape the editorial staff of 12 specialised agents that today (with the integration of Verify) is the heart of ProofPress and works on a continuous cycle monitoring some 4,000 sources on a global scale and selecting relevant information. The real innovation accompanying the platform, however, is not in the production of content, but in its certification. "ProofPress," Cinelli tells Sole24ore.com with conviction, "is the new frontier of doing journalism, because it makes the news true, transparent and fake-proof. The model, says the Italian manager, was born out of a precise need, that of countering the decline in reader confidence as the amount of accessible content increases exponentially. A paradox to which the platform born within the IdeaSmart ecosystem responds with the declared objective of transforming information into results that can be used without risk, precisely because they are accompanied by an automatic verification layer.

ProofPress' architecture traces the structure of a traditional editorial office, but distributes it over software agents: from source tracking to writing, from verification to publication, each phase is managed by autonomous components operating in parallel. The system performs multilevel content control by analysing thousands of sources in real time, cross-references data, verifies the identity of authors and produces a reliability report before publication. At the end of the process, each article receives a 'vignette' that allows the reader to trace the sources and method of verification. 'Today, anyone can produce content with AI,' Cinelli continues, 'but the real innovation is to create a system in which every piece of news is traceable, verifiable and cannot be altered. The ProofPress magazine itself, which you can register for free, represents an operational demonstration of the model: we are talking to all intents and purposes about an AI-native publication that produces verified content on a daily basis. The heart of the platform is, as mentioned, the Verify protocol, which introduces into the editorial cycle a step that is typical of digital notarization systems: the mechanism closely resembles that of the blockchain and is based on the association of a unique fingerprint to a given content, capable of guaranteeing its integrity and authenticity over time (any change, even minimal, generates a different hash, making any alteration immediately detectable). In the case of ProofPress, this principle is applied to the news, and the content processed by the agents is analysed and compared with certified sources subjected to a multilevel validation process and finally 'sealed' with an immutable cryptographic hash (and hence algorithm). The end result is content that is verifiable a posteriori and open to the reader, who can access the details of the sources and their validation process.

The question begs to be asked: does the editorial staff of AI agents drive the journalist out? The ProofPress model, despite its high degree of automation, does not exclude him, but rather repositions him by increasing his capabilities. The paradigm is that (widely debated) of the human in the loop and in this specific case this manifests itself in a human component that defines the editorial line, tone and reference sources and supervises the work of the artificial intelligence. 'The human being,' assures Cinelli, 'is always at the centre and controls and coordinates everything, while the agents execute, verify and certify'. The idea, set within IdeaSmart, is to combine scalability and quality control by creating the conditions for a single person to manage a production that, in a traditional context, would require a structured team. The platform, as Cinelli confirms, is currently being tested by two major Italian publishers and its use, not surprisingly, is linked to the possibility of expanding the content offering with new thematic verticals, without the need to build dedicated editorial offices. The other application front is the corporate one, with companies and organisations potentially interested in adopting the platform to create internal newsrooms, investor relations channels or vertical communication systems. The challenge is important and even provocative, playing on the measurability of trust: is the media market in Italia ready to accept and adopt a model whereby agent systems can not only produce content, but also verify it more quickly and systematically than traditional newsrooms?