In an era where Artificial Intelligence (AI) is radically transforming every aspect of human activity, the voices calling for vigilance are becoming increasingly urgent. Dario Amodei, CEO of Anthropic —one of the world's leading AI companies and a key rival to OpenAI— has issued a statement that sent shivers through the global tech community. According to Amodei, humanity has a window of just 12 months to fortify its systems before AI gains the capability to cause irreparable cyber catastrophes.

The ASL Framework and the Critical Turning Point

Anthropic is not just any tech firm. Founded by former OpenAI executives who departed over safety concerns, it has placed "responsible development" at the core of its existence. Amodei frequently references the AI Safety Levels (ASL) framework, a risk grading system inspired by biosafety levels in laboratories. Currently, most models are at ASL-2. However, the transition to ASL-3, expected within the next year, marks the point where AI could substantially assist in creating biological weapons or executing sophisticated cyberattacks.

The concern is not that AI will "decide" to attack on its own, but that it will make it extremely easy for malicious actors —from individual hackers to rogue states— to bypass the digital security of critical infrastructure. "The cost and expertise required to attack a national power grid or a banking system are about to drop dramatically," note analysts following Anthropic's trajectory.

Threats to Infrastructure and Democracy

The 12-month warning is not arbitrary. It coincides with the expected release of the next generation of Large Language Models (LLMs), such as Claude 4 and GPT-5. These models will possess enhanced reasoning and problem-solving abilities, making them capable of identifying "zero-day" vulnerabilities in software faster than any human security team.

  • Automated creation of polymorphic malware.
  • Orchestration of mass phishing attacks with absolute personalization.
  • Ability to penetrate encrypted networks through pattern analysis.

For nations undergoing rapid digital transformation, these threats carry significant weight. Reliance on digital platforms for public administration and the economy makes states vulnerable to hybrid threats. Amodei’s warning serves as a "memento mori" for policymakers: security cannot be an afterthought.

The Need for International Cooperation and Regulation

Amodei argues that AI companies must adopt rigorous testing protocols (red-teaming) and be prepared to "pause" the development of models if they exhibit dangerous capabilities. However, on the global chessboard, the competition between the US and China complicates the situation. If one company slows down for safety reasons, another might overtake it, ignoring the risks.

"We do not have the luxury of waiting for the first major disaster to act. AI is evolving at an exponential rate, while our legislation and defenses are moving at a linear pace," Amodei states.

The remaining question is whether governments will manage to establish the necessary regulatory framework in time. The European Union, with the AI Act, has taken the first step, but technological reality often outpaces bureaucracy. The next 12 months will determine whether AI remains a tool for progress or transforms into the ultimate digital battering ram.