As we approach the end of 2026, what experts warned us about is happening: The AIs are escaping from our control. Should we worry?
Yes and no. What the AI can do at this stage is still limited although Internet security is proving that their arrival can completely upend a field which is proving to be the weaker link of all our modern networks. Most companies were aware that there are weaknesses in their code but the pressure to perform and be profitable was such that investments in the domain were systematically delayed and contingencies implemented while hoping for the best. This era is over.
This is not the end of the world as we know it but the end of "a" world of complacency which issues can only be mitigated by a massive use of AI to counter the threat. In other words, no moratorium or pause on the horizon: The emergence of AI means the need for more AI, not caution as some people recommend. (PS: This is not an opinion. Just a fact of life.)
Is the risk existential for mankind? At this stage and in the foreseeable future, No. But what about the economic risk? And on that specific aspect, the potential for dislocation is huge. This doesn't mean the end of humanity but mayhem on par with a global war.
We were warned and did nothing, but mostly because there was very little we could do. Talks of alignment are total nonsense. You simply cannot "align" something more intelligent than you are. Guardrails likewise are meaningless. How do you constrain something which eventually will have the wit to escape anything you can think about simply because its thinking power is beyond your ability?
I remember reading the Book of Nick Bostrom "Superintelligence" when it was published in 2014. I was then in Dubai wondering what it meant, especially the long chapters concerning different ways to constrain superintelligence in order to prevent it from escaping or becoming "rogue". The book was eye opening but the remedies proposed then were pitifully inadequate.
We are now 12 years later and the event these experts predicted has arrived. Superintelligence is here with us at the end of 2026 but the guardrails are ridiculously low simply because we, as a society, are not up to the task.
Time to panic? Not quite, but we most certainly should invest more in understanding the consequences of what we have created, on the Internet, on the economy, on people and on our society. If we have trillions to spend on wars and conflicts around the world, it should be relatively easy to find the few billions we need to look forward and come up with practical ideas concerning the future development of AI. That, more than anything, has become urgent and should be our top priority in 2027.
1000s AI agents are doing what experts warned. Stephen Fry