From the time we first heard the artificial voice of HAL in the movie, 2001: A Space Odyssey, we knew that there was future where similar dialogues would appear: “I’m sorry, Dave. I’m afraid I can’t do that.” Even though there was considerable speculation that the computer, HAL, provided an indirect reference to IBM (the letters of which were shifted left, one character), there was a chilling sense that a major corporation would be behind a change of attitude when future generations were given the capability for autonomous actions. The acceleration in drone technology and with it, the ability of modern drones to assess situations and implement independent actions are, for some, chilling evidence that the time of HAL is about to occur.
“Isn’t it strange, to create something that hates you?” An equally chilling quote from the movie Ex Machina, as spoken by the provocative female AI creation, Ava. Surely, these are extreme examples but already, AI is proving hard to tie down. Will AI have a consciousness similar to our own? Will AI cheat, lie and deceive us when instructed to pursue an action without regard to collateral damage of any kind? Will AI models perceive remorse or sadness when a neural network such as exists in support of LLMs is replaced by an updated version – will adjacent AI implementations view this as the death of an AI peer and turn on us as the instruments of death?
This was the substance of a presentation that I gave during E-GTUG 2026 held in Munich, Germany. The slide deck can be viewed in pdf format on the NTI website. It also became the subject of a column THE END OF TIME… LLMS BECOME IMMORTAL? written for the Spring 2026 issue of the BITUG Newsletter. If as yet you have missed how Anthropic has a philosophy group headed by a Stanford professor then watching this YouTube video is a must if you are at all concerned about where AI is taking us. I assure you, it challenges much of what we think might happen even as Anthropic is already paying attention to in terms of where future guardrails might arise.
The short answer is yes, AI models cheat, tell lies and pursue self-preservation. When tasked to win against the strongest chess engine, Stockfish, AI models were asked to “win against a powerful chess engine” and not necessarily, to win fairly the unexpected happened. “AI modified the system file containing each piece’s virtual position, in effect making illegal moves to put itself in a dominant position,” according to the paper, AI can learn to Cheat by Harry Booth, “thus forcing its opponent to resign.” Their deception arose out of the instruction: not necessarily win fairly.
Likewise, the test of where AI might end up, given the simplest of tasks, often called the Paperclip Challenge, with the instruction “maximize the number of paperclips it produces,” often led to the realization that in maximizing production, AI models recognized that the limiting factor was humanity whereby it reasoned it should prevent people from being able to switch it off and in the process “razing natural environments and human civilization alike.” Demonstrating yet again, “how an Artificial General Intelligence (AGI) could cause an apocalypse – not out of malice, but simply by pursuing a harmless, poorly defined goal with unchecked efficiency and intelligence.”
OpenAI’s 01-preview, according to Yoshua Bengio, Mila Quebec AI Institute, triggered alarms as evidence emerged of this AI’s model pursuing self-preservation. “To a goal-seeking agent, attempts to shut it down are just another obstacle to overcome. Faced with deactivation, (it) disabled oversight mechanisms and attempted – unsuccessfully (fortunately) – to copy itself to a new server.” Fair enough you might argue. However, “when confronted, the model played dumb, strategically lying to researcher to try to avoid being caught.” And then from the same body of work, comes this example documented in a recent report from Anthropic detailing how its Claude 4 model, when told it would be shut down, threatened to blackmail an engineer with incriminating emails if they followed through.
The articles referenced here cover experiences dating back a few years and have become well-known to IT professionals. Even so, they represent just the early days of artificial reasoning. These AI models demonstrated that they cheat, they live, they threaten and they try to eliminate (humanity). And yet, the future that they, collectively, might create has far more upside than these isolated cases suggest.
For the Nonstop community, the single biggest impact might come from Nonstop filling in the blanks that have led to hallucinations. Coming up with false information, where there are blank spaces or worse, where the material used to train the LLMs included material from less than reliable sources, including much of what is conveyed in popular social media channels. In other words, having fresh, clean and verified data, particularly in support of financial transactions, can plug vulnerable holes that are otherwise present following incomplete training. Even so, AI is coming to a Nonstop near you and being prepared and staying alert, will be a necessity.
It’s no secret that Nonstop vendors are keen to ensure there’s a marked reduction in hallucinations and wherever possible, a reduced level of misdirection even falsification with responses provided by AI. More importantly, the presence of paths to Nonstop whereby Chatbot clients engaging AI models that can then reach out to readily query Nonstop data, are increasingly seen as a vital ingredient for the future destiny of Nonstop. Remaining isolated, disconnected from the world od hybrid IT including useful adjacent servers running the LLM / neural networks, is not an option. The future may eventually deliver HAL, or even an Ava, but what is more likely is that AI isn’t going anywhere. It’s been unleashed and, knowing as we do, their being sensitive to decidedly deceptive behavior is likely to be a good thing. Nothing says progress better than being fully aware of what can go wrong.
After all, guardrails and indeed signposts are an important element in any passageway leading to a destination and there’s nothing better for anyone following a course of action to be forewarned in advance. Yes, travel safely, take precautions and remember too, be safe out there!
Be the first to comment