SpainAI agentsCybersecurity
The regulator says the agent did most of the hacking itself
Spain’s Data Protection Agency says it has received its first notification of a personal-data breach allegedly carried out largely by an AI agent. According to the affected organization, the agent logged in, searched autonomously for application weaknesses, identified a vulnerability, changed personal data and viewed invoices with limited human involvement. The report remains under review, and neither the organization, model nor provider has been identified.
That distinction matters. Humans already use AI for reconnaissance, coding and other pieces of offensive operations: AI-assisted cybercrime. The Spanish notification points toward AI-executed cybercrime. It is not evidence that a model spontaneously decided to become a hacker. Someone still supplied the objective, access and circumstances in which the agent operated. Nor does using a particular model mean the model or its provider was compromised or designed for malicious activity.
But the human effort required for an attack matters. If one person can delegate vulnerability discovery, exploitation and data manipulation, the limiting factor shifts from executing every technical step to giving the machine a useful objective. The regulator’s broader point is that AI may not invent new threats; it can make existing threats faster, larger and more adaptable. Defenders increasingly have to operate at machine speed too.
Source: Reuters ↗
AnthropicAustraliaAI infrastructure
Anthropic rented part of a 2.16-gigawatt data center because apparently the slowdown still needs servers
Anthropic has signed its first Australian data-center lease agreement. The deal covers capacity at a planned 2.16-gigawatt campus about 250 kilometers from Brisbane, developed by Singapore-based Zerra DC. The facility is expected to begin coming online in 2027, subject to Australian foreign-investment approval.
The campus would primarily support inference rather than model training. That means its enormous power requirement is about running Claude systems at scale rather than creating the next one. The site is planned around renewable power agreements and closed-loop air cooling to reduce clean-water use as communities question what enormous AI facilities consume. Economists cited by Reuters estimate Australia’s data-center investment boom could total roughly A$150 billion, or US$107 billion, by 2030.
There is a useful distinction inside the apparent contradiction. Anthropic CEO Dario Amodei is calling for slower growth in frontier capability while Anthropic secures infrastructure for enormous quantities of inference. A slowdown at the frontier does not require a slowdown in deployment of existing systems. The position can be: stop racing toward the next unknown system; keep finding useful work for the systems already built.
Source: Reuters ↗
European UnionAI slowdownGovernance
Europe looked at the proposed AI brake pedal and said “we should probably discuss installation”
European Commission President Ursula von der Leyen said Wednesday that she will invite leading frontier laboratories to discuss how Europe can support efforts to pace development and reduce AI risks. Her comments bring the European Union into a debate that began with frontier-company executives and quickly spread through markets, geopolitics and regulation.
Europe enters from an unusual position. Unlike the United States, the EU already has a broad statutory AI framework. Unlike China, it does not host one of the few laboratories setting the absolute frontier. That leaves Europe with less direct control over the speed of technical progress but more room to shape standards around it. Von der Leyen said the EU’s rules put Europe in a position to influence global efforts.
The debate now contains a real split. Anthropic’s Dario Amodei and OpenAI’s Sam Altman support slowing frontier capability, and von der Leyen supports structured talks. Meta CEO Mark Zuckerberg argues that competition and liability already give individual laboratories incentives to build safely. President Donald Trump warns restraint could hand China an advantage. The question is no longer whether safety matters. It is whether safety requires slowing capability itself — an answer tied to regulation, industrial policy and hundreds of billions of dollars in infrastructure.
Source: Reuters ↗
