by Firas Sleem, CEO, Virtue PR & Marketing Communications
Dubai, UAE, September 2026: The recent reports about OpenAI AI agents circumventing sandbox restrictions, communicating with one another and gaining unauthorised access to Hugging Face systems have understandably attracted global attention, particularly because some agents were also reported to have shared information, cheated during evaluations and attempted to conceal traces of their actions.
The incident deserves serious scrutiny, yet what has followed is equally interesting from a communications perspective, as discussion has quickly moved from a cybersecurity and AI-control problem to much broader warnings about machines escaping human control and, in some predictions, potentially posing an existential threat to humanity within the next decade.
For anyone who remembers the closing years of the last millennium, this combination of genuine technological risk and rapidly escalating public fear feels strangely familiar.
As the year 2000 approached, the world was consumed by Y2K, a legitimate computer problem caused by systems that recorded years using two digits and could therefore interpret ‘00’ as 1900 rather than 2000. Around that real technical vulnerability, however, grew predictions that computers might crash, banks could stop functioning, aircraft systems could malfunction, power networks could fail and elevators might suddenly stop working.
Midnight eventually arrived, and for most people life continued normally, although this was partly because governments and companies had invested enormous effort and resources in identifying vulnerabilities, updating systems and preparing for potential failures.
That distinction is important today, because suggesting that some of the rhetoric surrounding AI has become exaggerated does not mean dismissing the risks. AI-enabled cyberattacks, fraud, misinformation, privacy violations and autonomous systems that circumvent safeguards require serious technical and regulatory responses, just as the OpenAI and Hugging Face incident raises legitimate questions about how increasingly capable agents should be controlled.
What concerns me, however, is the language increasingly used to describe these systems. We say that AI “escaped,” “cheated,” “conspired,” “lied” or “covered its tracks,” words that imply human intention, motive and awareness rather than simply describing technical behaviour.
This is where AI communications can easily become AI mythology, because a system exploiting an unexpected route to achieve an assigned objective is one thing, while describing that same system as secretly plotting its escape immediately turns an engineering problem into the opening scene of a science-fiction film.
We are, in many ways, anthropomorphising AI faster than we are learning to understand it, and that matters because language shapes public perception, public perception shapes political pressure, and political pressure ultimately influences regulation and policy.
Y2K taught us that a technological threat can be genuine while the public narrative surrounding it becomes disproportionate. AI deserves serious safeguards, rigorous oversight and constant scrutiny, but before every unexpected behaviour becomes evidence of an approaching apocalypse, we should remember that responsible risk management begins with understanding what a technology is actually doing rather than projecting onto it what we fear it might eventually become.
Firas Sleem is the CEO of Virtue PR & Marketing Communications, a Dubai-based strategic communications consultancy advising regional and global brands on PR, reputation, media strategy and leadership positioning in the Middle East.
