Intelligence services warn of next

For months, major technology companies have been engaged in a quiet but aggressive arms race to develop next-generation artificial intelligence models. While the public has grown accustomed to viewing these tools as advanced writing assistants or platforms for generating entertaining images and videos, global security agencies are watching the developments with increasing alarm. This concern moved beyond closed-door briefings this week, when cyber and intelligence agencies from the “Five Eyes” alliance — the United States, the United Kingdom, Canada, Australia and New Zealand — issued a rare and unusually sharp joint statement. According to the agencies, “advanced models capable of carrying out destructive and sophisticated cyberattacks against governments and corporations are only months away from deployment.”GalleryAnthropic CEO Dario Amoudi, right; OpenAI CEO Sam Altman(Photos: Getty Images, Reuters)Blocking Claude Fable and MythosAccording to a report in Britain's The Guardian, the immediate trigger for the alarming assessment was the U.S. government’s unexpected decision to block foreign nationals from accessing Anthropic’s latest model, Claude Fable. Alongside a more powerful version, Claude Mythos, the system is considered a breakthrough in its ability to identify complex vulnerabilities in computer systems.However, while earlier iterations were designed to serve as defensive tools for authorized companies, the new generation of models demonstrates a troubling leap in offensive capabilities. The ability to automatically generate exploit code from scratch, at unprecedented speed and scale, is fundamentally reshaping the rules of the digital battlefield.Security officials around the world emphasize that the industry is now at a historic turning point in cyber architecture. As computing power has grown and vast amounts of data have been fed into these systems, they have become increasingly adept at analyzing and understanding code. The ability to combine existing information, identify vulnerabilities and propose new solutions is what currently concerns cybersecurity experts most. What began a decade ago as modest academic experiments has gradually evolved into systems capable of writing, analyzing and improving code at unprecedented speed and scale.From consumer product to strategic assetAnd the issue is not limited to Anthropic. This surge in capability is also evident among competitors, including ChatGPT, Gemini and Meta’s open-source AI models. At the same time, China is advancing government-backed and commercial models under tight oversight from Beijing as part of its push for full technological sovereignty. In Europe, regulators are attempting to balance the strict AI Act framework with fears that the continent could fall behind in the technological and security race.Claude Fable 5 and Claude Mythos 5 were launched, and were almost immediately blocked(Photo: From the website of Anthropic)The U.S. decision to restrict access to advanced Anthropic models signals a deeper shift in how governments view artificial intelligence. If AI models were once seen primarily as commercial productivity tools, they are increasingly being treated as strategic assets with national security implications.Limiting foreign access to Anthropic’s most advanced systems reflects concern over the potential leakage of sophisticated cyber capabilities, and marks a transition into an era in which code and computing power are becoming part of the global balance of power between states.For intelligence chiefs, this is no longer a technical issue confined to IT managers, but a first-order national and economic risk. Not only has the timeline from model development to real-world deployment shrunk from years to mere months, these new tools are also lowering the barrier to entry for criminal organizations and rogue states, which can now launch thousands of sophisticated attacks in parallel without needing an army of human hackers.