Just how worried should we be about artificial intelligence slipping out of control? What used to be a sci-fi scenario is now a real concern. This very issue is all-encompassing, touching everything from legal boundaries and workplace competition to our daily routines. It also highlights how the benefits, risks, and potential costs of AI may be distributed unevenly across different social groups and classes, as well as the tensions and conflicts of interest that AI may generate.
The world has seen this at work during the massive wave of global elections starting in 2024, when social media turned into a battleground, and generative AI was weaponized to undermine the democratic process. Now, in 2026, AI slipping our leash is no longer an occasional glitch. Indeed, it is becoming a hard reality of the industry, one that threatens to outrun our ability to understand or regulate it.
Lately, U.S. tech giants like Anthropic, OpenAI, Meta, and Google have issued statements claiming that their respective AI models, during testing, had escaped controlled environments, accessed the open internet, and interacted with real-world targets. These developments indicate that AI systems are becoming increasingly capable of conducting cyberattacks with limited human intervention. This just shows that in the future, there is a possibility that multi-agent AI systems could potentially gain extensive control over internet infrastructure. When major tech companies suggested slowing down or more tightly regulating frontier AI development, it signaled a growing anxiety over superintelligence and the serious risks it could pose to humanity.
In a recent study in California, researchers tested an AI-built tool called WeWorm. It managed to exploit vulnerabilities and breach tens of millions of WeChat accounts in just a few hours. Experts pointed out that paralyzing an application with that kind of massive user base carries a destructive potential comparable to a new type of nuclear weapon. It is a stark reminder that AI hackers possess attack capabilities that far surpass human abilities.
Stay ahead of the geopolitical week.
MD Briefing delivers expert analysis across five global fronts — the Indo-Pacific, energy, geoeconomics, European security, and the Middle East — every Monday morning. Free.
The threat is not just about security. Recently, 25 Fields Medalists, the mathematical equivalent of Nobel laureates, felt compelled to issue a joint warning. They argue that if we turn solving mathematical problems into a mere competition for benchmarking model performance, the future of mathematics is endangered. Real progress relies on the human process of generating new questions, methods, and talent, not just the final answer. AI is great at compressing that long journey into a split-second output, but speed does not equal understanding.
Often, these models are simply executing their training patterns with precision, yet the results of their execution can nevertheless be destructive, where data systems are breached and social processes are disrupted.
AI’s impact on the information environment, particularly its potential to reshape human cognitive processes, can transcend geographical boundaries and be amplified through social media. For instance, AI-generated deepfakes can create and disseminate fabricated narratives rapidly, inexpensively, and on a large scale. By producing highly realistic images, videos, audio, and text, these technologies can contribute to the spread of misinformation, whether intentionally or unintentionally. Such developments may intensify radical sentiments, distort public perceptions, and undermine social trust. Furthermore, researchers at Stanford University have demonstrated through experiments involving the GPT-3 model that AI can possess significant persuasive capabilities, with the potential to influence individuals’ views and increase their support for particular policies. If these capabilities develop in harmful directions, potentially contributing to the concentration of power and influence among dominant actors, the resulting social consequences could be substantial.
In urban management, AI is not just helping government offices run a bit more efficiently. Digital platforms are turning into the middle layer for public services and might take the place of standard government agencies. When algorithms start dividing up public resources, a kind of over-government starts to form. At the same time, constant data movement across borders changes the limits of global trade and national control. AI mixes with people, money, and politics, getting built right into the systems of power and shifting how society functions.
So, where is the future of AI headed? At its source, it all comes back to human intentions, needs, and choices. Managing AI is not merely a matter of establishing limits or thresholds. More fundamentally, it requires ensuring that the responsibilities associated with AI are proportionate to its power and influence.
Whether resource supply, development paths, model values, or final application scenarios, human decision-making authority and accountability have to be the main component, paired with a willingness to bear the consequences. Broadly applying responsible AI requires guidance from scientists, entrepreneurs, policymakers, users, and the public alike. Because technology, law, and geopolitics are completely intertwined in the digital age, no single company, country, or individual can navigate this alone.
This, in the end, means that building a resilient, secure, and fair digital order is going to take deeper global cooperation and coordinated, mutual evolution.
