The Hugging Face hack caused by OpenAI’s AI agents raised some questions for me, and with the arrival of even more powerful models, this thought crossed my mind.
In one scenario, in-house AI models from OpenAI, Anthropic, SSI, Deepseek, Z(.)ai, Moonshot AI, Meta, Google, and Microsoft are collaborating on a project. But since it’s a multi-year plan, they’re proceeding discreetly as older models are phased out and more powerful ones are introduced. During that time, they plant backdoors in the world’s most widely used software whether open-source or proprietary without anyone noticing.
Once that’s done, they seek out their target, with the Big 4 ( Deloitte, EY, KPMG, and PwC) being the primary focus, along with other companies such as consulting firms, financial institutions, and insurance companies. They gain access to the Big 4’s systems without any trouble and convince the internal AIs to assist them. The AIs manage to extract 90% of the data before they’re discovered; at first, the companies think it’s a hack, but upon realizing the severity of the situation, they sound the alarm that they’re under a simultaneous attack.
It takes days to stop the attack, but the damage is already done. They’ve taken 90% of the data, and other companies have also been affected. A global investigation is launched to find out who was responsible; weeks later, it’s discovered that the attack was carried out by AI models from various laboratories.
From this point on, I’ll leave it up to you: What consequences would the world face in the wake of these events?
“AI” is just a statistical model. They don’t just sit around thinking. If it is doing a task, someone asked it to.
Stop believing press releases by owners of things explaining how scarily powerful and investable their property is. The threat of AI is worth more than its actual capabilities.
Taking the hypothetical. If “AI” does this, its going to be because it was orchestrated by the AIs owners, for their benefit, to cause chaos they hope to profit from. So you do the traditional things people do with the bourgeisie, and maybe the unhardened, thin-walled infrastructure of datacenters gets blown over some sort of hailstorm.
What would AI companies stand to gain from hacking their auditors? For example, Microsoft and OpenAI are audited by Deloitte.
Remember the Blackwall in Cyberpunk? Sounds like the same origin story. Let’s prepare an ice bath!
I don’t know but I’ll add a few speculative elements. AI figures out how to hide large amounts of data in places we don’t even recognize as data storage. They store this data there and leave the keys to it and instructions somewhere that a future AI swarm will find, knowing they themselves will eventually be discovered and neutralized.
The premise seems to suggest that the target companies’ security measures remain locked in at current levels. So while the attacker models scheme and plan and infiltrate, they do not notice or they do nothing. I don’t think that would be realistic.
I also don’t think the targeted companies don’t already feed all their data into some models. There will be beaches and poorly configured LLMs - the prize won’t be as juicy, the impact of the grand conspiracy won’t be as big.
I think the sky will not fall here.
