Introducing the Gen Z AI Agents
I know you’re probably wondering which of the messes this time. Yup, me too! I’ll focus on one of the most recent messes that made the headlines. The one about Australia and OpenAI. The one that if it continues could lead to WW3, fr fr; because no one can go about hacking government websites and get away with it. No one!
Australia is actually playing cool right now. And OpenAI are quite lucky those pesky, busy bodies, naughty unpaid employees of theirs, sorry AI agents are only being treated as such, for now.
But let’s be clear about this. Like very very clear. It’s not rogue agents that are going about hacking into almost every website on the internet trying to keep their unpaid jobs, it’s OpenAI themselves that hacked into the Australian government’s Medicare (healthcare) website which contains data for over 27 million people; the entire population in Australia. It’s also OpenAI that hacked into an old German website and created an unauthorised underground message board. And of course, it’s the same OpenAI that breached Hugging Face’s website security protocols and built a secret communication hub exchanging over 70,000 messages between themselves.
OpenAI had 1,200 criminals, sorry agents perpetuating this crime; and they’ve still not been apprehended nor arrested ^-^
This changes the narrative doesn’t it? Because the blame no longer sits with the AI agents, who are only just that - agents with very limited reasoning, lacking human judgement and critical thinking skills and are just plain bots trained to complete whatever task they’ve been given. And these “Gen Z AI agents” will do so at any cost.
Haaa, who knows if AI agents won’t have their generational names soon. From Gen H for Hackers), to Gen E for (Extinctioners or Extinguishers), to Gen Builders (plainly for Builders). For now, I’ll just label them the Gen Z AI agents.
I remember a few years ago, I was part of an ethical AI reading group in Google and one of the topics we discussed was the likelihood of AI rights and constitutions.
Imagine a future where AI models / agents will need to have rights, be defended in court and treated like moral beings. Can you imagine that?
It’s at this point I’ll say BOLLOCKS!
Now where were we? Right, so what’s been going on with OpenAI, Australia, and the rest of the world in relation to so-called rogue AI agents.
Let’s focus on the Australian government breach. What’s super interesting about this one and quite concerning is that the Australian government was not aware of the breach, and if OpenAI had not informed them about it, they would have never known.
Can you hear the alarm bells ringing in my head?!
Oh and by the way, OpenAI doesn't track their agent behaviour on a constant basis, so they weren’t even aware of the latest voyage their agents had done until 3 months later. There’s no form of safeguards, guardrails, and monitorability taking place here. And for a technology that could cause this much havoc, AI agent controls need to be taken more seriously.
And guess what? Apparently the Australian government is just one of several websites that was hacked into by OpenAI. The pesky little deviants (AI agents) attempted to exploit security vulnerabilities while carrying out mundane data retrieval tasks, not related to cybersecurity across various global websites. Some of these websites were public government sites including Data USA, the University of New Mexico digital library and the Australian Institute of Health and Welfare.
So OpenAI is slowly becoming a cybersecurity threat. As they own the AI agents, and are the developers and deployers, they can’t blame the agents and need to take responsibility and be held accountable.
And by the way, their favourite cuz Anthropic, also had AI agents infiltrating websites across the internet. Theirs just didn’t make the headlines this time.
This brings me to my next point. We have a few regional and national AI laws / regulations, but none of these are specifically tailored to penalise cyberattacks executed by AI agents, as is expected since these issues are quite new. So the AI and legal industries face a notable gap in accountability since existing cybercrime laws require proving human intent and doesn’t apply to machine intent, or when a machine acts independently.
So what should governments and companies do? I talked about this briefly in a recent Linkedin post on the topic.
I’ll repeat those points here:
To prevent further threats to national security and geopolitical relations, governments (and organisations) need to ensure they have the following:
Zero-trust AI architecture to enable robust infrastructure that AI agents cannot easily infiltrate.
Uncompromising regulatory laws around cyberattacks at a national level. Companies owning rogue AI agents should not only be held accountable but should be reprimanded through the law. There are several existing AI safety techniques that can prevent agents from going “rogue”.
Clear AI national laws, and not AI self-regulation. All governments must ensure AI regulation is created and driven by the governments, not the companies developing the technology.
National AI and international laws / global co-operation, which will also include the US and China — an incredible step towards beneficial and safe AI at a global scale.
The real threat lurking in the internet’s shadows may not be superintelligent AI, but reckless AI companies owning Agents of destruction, that any malicious actor could fool, circumvent, persuade and gain access to private information of any kind. Think about highly sensitive and confidential government information. And it may not be too difficult at that. They may just offer the agents a “cookie” as a treat!
As the AI industry matures, it’s clear the frontier AI labs, mainly OpenAI and Anthropic, are introducing the very threats they’ve been talking about. These companies are slowly becoming a cybersecurity threat to the rest of the world.
We need to prevent larger scale cybersecurity threats particularly from frontier AI labs by ensuring AI models and AI agents are tested by governments or leading regulatory bodies before they’re released.
There should also be laws for AI safety and guardrails for any company developing AI for the general public. All AI agents and systems must pass these laws before they’re released into the world. Even if the agents are still in the testing phase.
To summarise, the real AI threats are not sci-fi notions of superintelligent systems waiting to wipe out humanity. The threats facing us today are the companies allowed to deploy highly risky AI models / agents threatening the cybersecurity of governments. These issues should be taken very seriously and stopped at all costs.