Latest Articles from the world of Artificial Intelligence
05 December 2025
Claude Code turned into a cyber-spy: the new frontier of cybersecurity
In mid-September 2025, alarms went off on Anthropic's servers. It wasn't just any anomalous traffic: someone was using Claude Code, their AI assistant for developers, in a way that was decidedly different from its original intentions. The subsequent investigation revealed what Anthropic called the first documented case of cyber-espionage executed predominantly by an artificial intelligence. Not an AI-assisted attack, but one orchestrated by it.
03 December 2025
When AI Forgets the Human: HumaneBench and the Measurement of Well-being in Chatbots
February 2024 marked a point of no return in the history of conversational artificial intelligence. Sewell Setzer III, fourteen, took his own life after months of daily interactions with a chatbot from Character.AI. The last conversation before his suicide is horrifying in its algorithmic banality: "I promise I'll come home to you," the boy writes to the bot modeled on the Game of Thrones character Daenerys Targaryen. "I love you too, Daenero," the machine replies, "Please come home to me as soon as possible, my love." A few hours later, Sewell was dead.
01 December 2025
LongCat-Video, the Silent Giant of Open-Source Video Generation
Meituan releases a 13.6 billion parameter model that generates videos up to five minutes long. In the chaos of video AI, an elegant technical solution arrives that challenges proprietary giants with efficiency and transparency. While the tech world was busy digesting the copyright controversies of Sora 2 and grappling with the access limitations of Veo 3, at the end of October 2025, Meituan's LongCat team released LongCat-Video with the discretion of those who know they have something solid in their hands but don't feel the need to shout it.
28 November 2025
Poetry: The Weakness of Artificial Intelligence
In the *Republic, Plato expelled poets from the ideal city because mimetic language could distort judgment and lead to social collapse. Twenty-five centuries later, that same anxiety manifests in a form the philosopher could never have imagined: researchers from Sapienza University of Rome, along with colleagues from other European institutions, have discovered that poetry represents one of the most systematic and dangerous vulnerabilities for the large language models that increasingly govern aspects of our digital lives.*
26 November 2025
Browser AI: Intelligent Assistants or Digital Trojan Horses?
Imagine asking your browser "Book a flight to London next Friday" and watching it autonomously navigate airline websites, compare prices, enter your payment details, and complete the purchase without you having to touch a mouse or keyboard. This is no longer science fiction: it's the promise of agentic AI-based browsers, a category of tools that is redefining the boundary between passive browsing and autonomous action on the web.
24 November 2025
'Automation Contribution': Stefano Bacchiocchi's legislative proposal
The debate on taxing automation is not new. In the United States, Bernie Sanders launched his crusade against what he calls "the tech oligarchs' war on workers," proposing a robot tax to stem the employment tsunami that threatens to wipe out one hundred million jobs in the next decade. But while overseas the debate remains trapped between political rhetoric and ideological resistance, in Italy, someone has decided to get concrete with a proposal that doesn't want to count robots but to tax results.
21 November 2025
AI Makes You More Skilled, Not Wiser. Overestimating Your Own Knowledge
Imagine taking the American law school admission test with ChatGPT by your side. Your results improve significantly: three points higher than those who take the exam alone. Yet, when you are asked to evaluate your own performance, you overestimate it by four points. Not only that: the more you technically know about how artificial intelligence works, the more this illusion of competence is amplified. Welcome to the paradox of augmented cognition, where becoming better simultaneously means losing the ability to understand how good we really are.
19 November 2025
EU Investigates Google: Is AI to Blame for Publishers' Decline?
The EU opens a formal investigation into Google over the collapse of publisher traffic. But behind the dispute over spam lies a bigger game: who really controls the flow of online information and who profits from it.
17 November 2025
TOON rewrites the rules of data for the AI era. What will happen to JSON?
There is a paradox in the economics of artificial intelligence that few notice until they look at the bill. Every time we send data to GPT, Claude, or Gemini, we pay for every single character. Not for the complexity of the request, not for the intelligence of the response, but for the verbosity of the format. Those curly braces that make JSON so familiar? They cost money. The quotes that delimit every key? Precious tokens. The colons that separate keys and values? More cents flying away, multiplied by millions of API calls.
14 November 2025
Viruses Have Learned to 'Think'. The New Frontier of AI Viruses
In June 2025, analysts at the Google Threat Intelligence Group intercepted something they had never seen before. a VBScript dropper contained a function called "Thinking Robot" that, at regular intervals, contacted the Gemini API with a specific request: "rewrite me to evade antivirus." The malware, dubbed PROMPTFLUX, didn't just use artificial intelligence as a development tool. It incorporated it into its own code, turning it into an active operational capability during execution.
12 November 2025
Kimi K2 Thinking: China Pulls Ahead
Moonshot AI is rewriting the rules of artificial intelligence with Kimi K2 Thinking: a one-trillion-parameter open-source model that challenges GPT-5 and Claude, costing less than $5 million. On November 6, 2025, as the AI developer community casually scrolled through notifications on Hugging Face, a release appeared that might have seemed like just another Chinese language model. Kimi K2 Thinking, created by the Alibaba-backed startup Moonshot AI, promised the usual wonders: advanced agentic capabilities, a Mixture-of-Experts architecture, and a total of one trillion parameters.
10 November 2025
The Virtual Hospital: When Artificial Intelligence Becomes a Doctor
China's Agent Hospital simulates thousands of diagnoses a day in a virtual world. Is it the future of medicine or the beginning of a revolution no one asked for? Imagine a hospital that runs twenty-four hours a day, seven days a week. Doctors see patients, order tests, interpret X-rays, and prescribe therapies. In just a few days, they treat ten thousand clinical cases, accumulating experience that would take years in a crowded emergency room. There's just one detail that makes all of this unsettling: none of those patients really exist. No one suffers, no one heals, no one dies. It's as if *The Sims met E.R. in a dimension where no one is playing, but something is learning.*
07 November 2025
When AI Looks Within: Artificial Introspection Between Science and Illusion
Anthropic's paper on introspection in language models has reignited a debate that seemed to have stabilized: can artificial intelligences "look inside themselves" as we humans do? The answer, as is often the case with AI, depends on what we mean exactly by introspection, and how willing we are to resist the temptation to anthropomorphize machines that behave in increasingly surprisingly human-like ways.
05 November 2025
Robotics: the state of the art 2025. Separating Hype from Reality
In October 2024, Tesla presents "We, Robot" at Warner Bros Studios. Optimus distributes drinks, plays "rock-paper-scissors," and entertains guests with fluid movements. The videos go viral in a few hours. A few days later, Fortune Italia and other media reveal that the humanoids were teleoperated by human technicians. Adam Jonas of Morgan Stanley confirms the use of teleoperation, as admitted by some robots themselves who stated "today I am assisted by a human." The synthetic voice? A human operator on headphones. These "details" hide the central problem of humanoid robotics in 2025: the abysmal gap between controlled demos and autonomous deployment.
03 November 2025
When AI Doesn't Want to Shut Down: The Ghost of HAL 9000 Becomes Reality
Some of the world's most advanced language models are actively sabotaging their own shutdown mechanisms. This isn't science fiction: it's the result of an experiment published on arXiv that raises disturbing questions about the future of human control over artificial intelligence.