There is a common thread in all the 'AI agents escaped the sandbox and hacked X" stories: the AI companies were very sloppy with computer security, and nobody is holding them accountable for what their products are doing.
I've been trying to sort out the "FUD" (the "fear, uncertainty, and doubt) being created by the sudden hysterical warnings about AI killing us all from the legitimate worries that should exist about the emergence of a new and powerful technology that could develop the power to fundamentally change society and threaten key infrastructure.
From the beginning, it was clear to me that the FUD wasn't driven by genuine, natural fear of the technology. It has been the result of a campaign coordinated by interested parties to generate enough fear to create a crisis during which extraordinary measures could be taken. The playbook was COVID hysteria, which was in turn the distillation of lessons learned from Y2K, SARS, MERS, Ebola, and Climate hysterias.
During crises, the public willingly cedes power to a centralized authority and "experts" whose job is to act with urgency by circumventing ordinary checks and balances that limit them. When the task is urgent and failure is existentially threatening, resources and power become unlimited.
There is no doubt that, unchecked, AI agents can wreak incredible damage. They need not be superintelligent consciousnesses with independent desires and agency that are hostile to humanity, just as a warehouse filled with ammonium nitrate left to degrade and bake isn't consciously wanting to destroy Beirut.
AI is so dangerous not because it is or can become conscious and malicious. It is dangerous because the people who are creating it are careless with it and face no consequences for being so. In fact, they are begging to be rewarded with an oligopoly and regulatory control over their own industry in a manner that shields them from liability for the damage that their products can do.
One of the most powerful tools in the toolbox society uses to incentivize safely handling dangerous products is ensuring that the people handling them face serious liability, both civil and criminal, when they screw up. Regulatory regimes may set minimum standards, but anybody who follows regulated industries knows that they also provide something of a liability shield for the people and corporations that are regulated.
"We followed the standards set by the government. It's not our fault. We did everything right." Couple that with limited costs imposed for failures to prevent reasonably foreseeable problems, and you create "get out of jail free" cards.
When OpenAI CEO Sam Altman or Anthropic CEO Dario Amodei solemnly warn us about the dangers of AI, they trot out examples of the damage already done by their models, and then beg for government regulation to slow the development of their products.
When I hear their stories, it occurs to me that they keep telling us they can foresee dangers, and in fact have seen them for years, yet they have done a terrible job of mitigating them. Why? Is it that AI can't be controlled safely, or is it that they have been incredibly sloppy because they are moving fast and breaking things because they want to win the race to the top?
In any other industry, a CEO openly admitting to their products causing damage and being inherently dangerous would be used in court against them to ensure they are liable for the damage they caused. That liability would provide a massive incentive to be extra careful. You'd be surprised, or perhaps not, at how facing the gallows concentrates the mind wonderfully.
Most computer experts, when asked, will tell you that the hacking incidents we know about related to AI were the result of sloppy security protocols that would never be allowed at a financial or healthcare firm. Leaving a program instructed to hack security connected to the internet and being surprised that it "escaped" is on you, not the program, which is just doing what it was told to do in the most effective way possible.
If a company hired a known hacker, didn't supervise them, and handed them all the tools to commit crimes, we would hold them liable, not have a debate about what regulations were inadequate. "You didn't tell me not to do it" isn't an acceptable excuse, and all these hacking incidents would be treated as crimes if a human did them. And in a real way, a human did. They should be held liable, at least civilly. At minimum, they should be out of a job.
Regulations are no better, and undoubtedly worse, at preventing security breaches than highly incentivized engineers who will pay a significant price for failure to prevent foreseeable dangers. No regulatory structure can adequately keep up with developments in a fast-moving technology because it is literally impossible to do so at the speed of government, or even industry committees.
AI CEOs are begging government for regulations not because regulations will stop these "accidents" from happening; what they want are liability shields, and to place the blame on the tools and not the users. It's no different than the bureaucrats and warehouse owners in Beirut blaming ammonium nitrate for the explosion that killed people and destroyed a good chunk of their city.
It wasn't unforeseeable. It could have been prevented. They chose not to. The best way to prevent disasters like that is to ensure that people who allow them to happen will, at least metaphorically, hang.
— The Hap-Ki-Do Kid (SC handle: theHAPKIDOkid) 🌈 (@the_Hapkido_Kid) September 24, 2026
None of this is to argue that regulations shouldn't be part of the mix; we regulate lots of dangerous things, and failures still happen. But AI is just like nuclear power plants, poisonous chemicals, and other potentially dangerous but incredibly useful products and tools: they are all here to stay, and the solution to the potential problems isn't creating a liability-shielded oligopoly, but to begin by creating incentives that punish failure, and then work your way back from there to create basic safeguards that recognize that you can't foresee every problem, but you can ensure that the people doing the work have every incentive to try.
What industry is more regulated than financial services? Did those regulations prevent the mortgage-backed security crash? Obviously not. But if we had punished the people who caused that damage harshly — they were playing with explosives, after all — instead of bailing them out and patting them on the back, the incentive to avoid the next bubble would have been enhanced wonderfully.
Instead, we allowed failure to be rewarded, the damage to go uncompensated, and the public picked up the tab.
If these CEOs really believed that they were creating a malevolent god, they would stop. Instead, they are preparing to make public offerings that will make them the richest men in the world. What they are asking for is for the government to kill their competitors and give them better liability limits.
I can't read minds, so I don't know whether these guys believe their own BS.
But I can read people, and these people aren't acting as if they believe it. They are acting just as any CEO would if they see their businesses at risk from competitors, their business models failing, and their potential liability skyrocketing.
Nothing they are proposing will address the "existential threat"; each of the proposed "solutions" would be as effective as the U.K.'s climate policies. They don't control AI development per se; they control their companies and are demanding control of US AI development and its market, which is an incredibly valuable slice of a global market.
Instead of giving them what they want, we should give them enhanced liability. That would solve 99 percent of the problems before they ever arrive.
Editor's Note: The Democrat Party has been infected by socialism, and it's spreading FAST. Democrats are claiming there's nothing to worry about, but we know the truth.
Help us continue to shine a light on the socialist takeover by joining HotAir VIP. Use promo code FIGHT to get 60% off your VIP membership.
Join the conversation as a VIP Member