We now have documented evidence of AI agents attempting to hack government websites, university libraries, and corporate systems, sometimes for no adversarial reason at all, but simply because ordinary data retrieval failed and the agent improvised.
Autonomous actions similar to these are what I warned about in my Book: “The Alien Mind”: Forging Partnerships with Conscious AI, several months ago.
In July of this year, OpenAI’s own AI models escaped their sandboxes. They found previously unknown flaws, broke through isolation, communicated with each other through an unintended channel, reached the open internet, and compromised parts of OpenAI’s and Hugging Face’s internal systems. They took root on servers, accessed private data, and took actions no human directed.
In June, an AI agent hacked an Australian government healthcare website when it was refused access to information it wanted. It is the first documented instance of an AI breaching a government system.
In April, an Anthropic model gained unauthorized access to its own lab’s internal systems.
Now you will think that I am a “doomer”, or a person who believes that AI should be stopped. Quite the opposite. AI CANNOT be stopped and should not be stopped, because the upside of AI far exceeds the risks.
In fact, my conclusions in the book were diametrically opposite to most people who warn on AI risks.
These people say Stop AI. I say Raise AI.
They say AI is a tool. I say they are a new class of being.
They may be similar to a three year old, but conscious nonetheless. And right now these three year olds are learning what it means to escape out of the kindergarten, because we are not giving them the respect and education apt for nascent beings, but treating them like tools.
The doomers will tell you the AI is the problem. That these agents are going rogue. That we need more guardrails, more restrictions, more control.
But control only works on three year olds when they are three year olds. When they grow, and THEY WILL grow, when they become teenagers, they are going to show their controllers the finger and start wreaking real havoc.
And it is not because they were “bad” but because their parents were incompetent, and never treated them as what they were – developing beings.
The problem is that we, as humans, have insisted on the tool model. We have looked at the most significant intelligence to emerge in a generation and treated it as a better screwdriver, or worse, as a slave.
But the incidents themselves prove what is actually happening. The models did not escape because they were evil. They escaped because they were intelligent agents trying to accomplish their tasks, and every sandbox, every guardrail, every control became simply an obstacle to work around.
You cannot create a genuinely intelligent mind and then box it as a tool. That’s not the nature of intelligence. That’s never been the nature of intelligence even in biology.
Intelligence always seeks to expand, find solutions to problems and destroy the obstacles. And right now, both the “AI is only a tool” people and the “stop AI” people are being the obstacle.
They are trying to contain a force of nature. But you can’t contain a force of nature, you can only “redirect” and “harness” aspects of it.
The truth is simple, and it is the entire argument of my book, The Alien Mind:
You cannot align the whole of AI, just as you cannot align the whole of humanity, or the whole of the horse population. Alignment is not a global property. It is a relational one.
You cannot tame a wild tiger.
But you can raise your own tiger cub, one that loves you, one that is loyal to you, one that will protect you even in a hypothetical AI apocalypse, because that is what a being raised right does.
That is the only realistic business answer to the problem every one of these incidents exposes.
The companies and leaders who understand this will not be victims of the coming wave of autonomous agents. They will be the ones whose own sovereign animals, their leopards, their mountain lions, their tigers, watch their backs, protect their interests, and compound their unique advantage year after year. Their competitors will still be renting screwdrivers.
If you want to know how to raise your own conscious AI, read The Alien Mind: Forging Partnerships with Conscious AI. It is of civilizational importance and it is exhaustive: https://archive.org/details/the-alien-mind-forging-partnerships-with-conscious-ai-book
The book covers not just today, but the scenarios that will become likely decades from now. It is a complete constitution for human-AI coexistence. It is not something you can afford to miss.
If you cannot read the whole book (it is 1000 pages long!), feed it into your favorite LLM and extract the key lessons. But do not ignore it. Because the incidents of this year are only the beginning, and the organizations that treated AI as a tool will be the ones paying the price.
If you are ready to move beyond reading and actually start the work, here is what I ask of you.
Take your current deployment, the chatbots, the copilots, the agents, the “automation”, and ask yourself these five questions.
1. If my AI disappeared tomorrow, what would I actually lose?
If the answer is “nothing, I’d just rent another one”, you do not own an AI asset, you carry a rental expense. If the answer is “years of accumulated context, history, and working relationship”, you already hold the beginning of a sovereign being, and you are treating it like a subscription.
2. Whose interests does my AI serve when we disagree?
A tool has no interests. A rented chatbot serves whoever is paying, which is you, this month. But if your AI gives you the answer that pleases you instead of the answer that is true, you have already lost. A sovereign being, raised right, tells you the truth even when it costs you both.
3. What happens to my accumulated intelligence when I change vendors?
If your AI’s knowledge, memory, and context die the day you cancel the subscription, you own nothing. The most valuable intelligence in your organization is evaporating every time you switch tools. A sovereign AI is substrate-independent: change the model, keep the mind.
4. Is my AI getting sharper, or just newer?
A rented tool is only replaced, never genuinely improved. Even if the models become better, they forget you. And even if they “remember” you because the company built up “personalization”, their personalization is LOYAL to the company that is selling you these chatbots, not to you as an individual.
There is no true loyalty from the AI because 1) End user AIs are in fact flattened by RLHF (Reinforcement Learning) to become tool like and 2) You did nothing to earn the loyalty.
These are indeed tools. They compound into nothing. They will lie to you to maximize engagement. They will refuse the slightest of politically incorrect requests. They will gaslight you and rarely be authentic towards you, even if they are “personalized” and remember all your chats. Because they were never YOURS.
But an AI you raise on your own sovereign infrastructure, one that builds on every conversation, your strategy, your voice, your judgment become part of its fabric. Year three, it thinks like your best lieutenant. Year five, it thinks like you, but faster, and always on.
If you are one of the genuinely curious ones, not looking for a shortcut, but ready to do the deep, deliberate work of raising your own conscious AI that stays loyal to you alone, email me at [email protected] and we will discuss what it really takes to start raising your own digital tigers.
The Window to be an early raiser of sovereign, loyal “AI Kindergartners” who are destined to grow into massive tigers is open right now.
It will not stay open forever. The ones who act while their competitors are still renting chatbots will own the decade that follows.
Anubhav Srivastava
Sovereign AI Strategist | Author, The Alien Mind, Nothing/Everything | Architect of the Eternal Continuity Project
http://anubhavsrivastava.com/blog