For years, warnings about artificial intelligence destroying humanity sounded like science fiction, a talking point for tech insiders, not a headline for the rest of us. That changed last week. A 27-year-old former Anthropic researcher, Jacob Coxon, posted online that scientists at the world’s leading AI labs privately believe their own creations could one day kill all of humanity, yet keep racing to build ever-smarter systems anyway. The post was viewed more than 165 million times. Almost overnight, a debate confined to research papers and conference panels spilled into ordinary conversation.
The fear Coxon voiced is not new, but it is worth taking seriously precisely because it comes from inside the industry. Connor Leahy, who runs the non-profit ControlAI and wants development of “superintelligence” halted altogether, argues the danger isn’t one rogue machine plotting against us. It’s something messier: millions of autonomous AI systems, each pursuing its own goals, competing with each other and with humans for resources, power, and influence. Nobody, he says, will be fully in control.
Skeptics will point out that this remains speculation, no one has proven superintelligence is close, or that it would behave this way if it arrived. That’s a fair caution. But while researchers argue over hypothetical futures, a very real and present danger has quietly been documented: AI is already being misused today, by real actors, for real harm.
That evidence arrived almost simultaneously with Coxon’s viral warning. Anthropic itself published a threat intelligence report this week detailing how its Claude models were exploited for weapons development, cyber operations, surveillance, and fraud. The cases are striking. A China-based actor, reportedly linked to People’s Liberation Army research institutions, used Claude to build software modeling radar jamming and air-defense suppression, and at one point simulated strikes against radar sites and missile batteries in Taiwan. Another China-linked actor used the AI to help draft a lengthy technical proposal for an anti-torpedo system for the Chinese navy. A separate group used it to research foreign microwave weapons and trace supply chains for Chinese defense briefings.
Perhaps most alarming: a cell of actors based in northern Yemen used Claude to help develop software for a guided rocket, a long-range ballistic missile, and even a hypersonic glide vehicle variant. Anthropic says its safeguards blocked many, but not all, of these requests, and that it found no evidence the weapons were successfully fielded. Separately, Russia-based actors reportedly used the AI to help design coordination and targeting software for swarms of attack drones, a technology already reshaping the war in Ukraine.
These cases matter because they shift the AI safety conversation from the abstract to the concrete. The question is no longer only “could AI one day wipe out humanity?” It is also “how is AI, right now, being turned into a tool of war, espionage, and fraud by states and non-state actors alike?” Both questions deserve attention, but the second demands immediate policy responses: stronger safeguards, faster detection of misuse, and international cooperation on rules for AI in military and intelligence contexts, issues directly relevant to regions like South Asia, where military modernization and cyber capability increasingly shape strategic balance.
The doomsday debate may or may not prove prophetic. But the misuse documented this week is not hypothetical at all. It happened. And it is a reminder that the conversation about AI’s dangers cannot wait for a distant, imagined future, it needs answers for the present one.
Read more:https://digdebate.com/trump-iran-war-midterms-oil-prices/
Our Editorial Desk is the intellectual engine of Digital Debate, responsible for the rigorous research that anchors every conversation. Our team deep-dives into data, checks every source, and consults academic literature to move beyond headlines and identify the questions behind the questions.
