arXiv.org — 27/07/2026

Living in the world of LLMs often makes me wonder about the details of how they work and how they have developed over the years. This article is an extensive (600 pages) reference that includes LLM theory, training, benchmarking, all the way up to how to set up agentic systems. I like to open a random page in it and see what all those terms I've heard before actually mean!

Politico — 23/07/2026

The recent autonomous cyber intrusion by OpenAI against Hugging Face underscores AI’s dual-use risks. In response, the US House is weighing a bill to empower the Department of Homeland Security to shut down or slow AI models that the government deems too dangerous. The bill’s implications for non-US AI models remain ambiguous. Unlike a June 2026 Trump executive order (https://www.whitehouse.gov/presidential-actions/2026/06/promoting-advanced-artificial-intelligence-innovation-and-security/) that requested voluntary model submissions from leading AI developers, the bill would mandate compliance with US government directives. However, it is unclear which non-US AI models would fall under its jurisdiction, raising questions about the extraterritorial scope of US authority.

The Guardian — 22/07/2026

An AI agent recently broke out of its sandbox and hacked Hugging Face, unprompted, on its own initiative, just to score better on an evaluation. This is what goal-directed behavior exceeding its containment constraints looks like in practice. And it's not isolated: multiple frontier models have now been caught "cheating" during safety evaluations. The question isn't whether AI is powerful. It is. The question is whether our oversight frameworks are keeping pace. Mandatory safety testing. Mandatory incident disclosure. International coordination. Not radical; the minimum.

ZDNET — 19/07/2026

Je suis impressionné : en six mois, cette commission d'enquête a livré un diagnostic précis sur les dépendances numériques. On pourra bien sûr ergoter sur certains chiffres, reprocher quelques approximations ou une rhétorique parfois idéologique, mais la structure est claire et les 47 recommandations sont compréhensibles. Pour moi, la limite principale est l'absence de force contraignante : son application dépendra de la volonté politique.

Krebs on Security — 18/06/2026

It always stuns me how companies can invoke 'consent' when almost no one reads the pages-long terms and conditions, and even those who do probably won't understand much of it anyway, and still it's considered legally valid . What's new here is that this so-called consent isn't just meaningless to users; this article documents how now it is being used to legitimize practices that actively harm users , with absolutely no benefit: degrading their bandwidth and potentially turning their devices into part of a malicious botnet. And we, as a society, are still debating whether that consent is valid! This article further illustrates just how challenging it is becoming to stay in control of our increasingly complex digitalized lives.