AI represents one of the largest technological leaps I have witnessed in nearly 30 years in technology and cybersecurity. These models are advancing faster than anything we have seen in modern technological history.
The central concern is that this is fundamentally different from traditional software development. With conventional programming, developers explicitly define how a system should behave.
Advanced AI models can be difficult to predict and constrain, and they can demonstrate capabilities their creators did not fully anticipate. As disclosures from companies such as OpenAI and Anthropic have shown, these systems are already capable of performing sophisticated cyber operations against real organizations.
Dario Amodei and employees at several frontier AI companies have called for slowing development long enough to establish meaningful safeguards around this extraordinarily powerful technology. I understand and share many of those concerns. The geopolitical problem, however, is that the United States currently holds an advantage over China. If we slow down unilaterally, we could give China the opportunity to close that gap, surpass us, and fundamentally shift the global balance of power. AI has the potential to be that consequential.
We are entering an extraordinary period.
Over the next several months and years, AI could help us solve mathematical problems that have eluded humans, advance quantum physics, discover treatments for devastating diseases, and accelerate scientific progress across nearly every field. That transformation is not theoretical.
It is already underway.
I believe AI represents the next nuclear arms race. The crucial difference is that much of the power is currently concentrated in private companies rather than governments. That makes this moment fundamentally different from the Manhattan Project.
There is also an economic reality that cannot be ignored. Frontier models reportedly cost enormous amounts of money to develop and may remain state of the art for only a few months before the next generation replaces them.
That pace raises serious questions about sustainability, revenue, and widespread adoption. Meanwhile, Chinese AI companies have become highly effective at using techniques such as distillation to reproduce capabilities from leading models at a fraction of the cost.
Depending on how that knowledge is obtained, this can raise major intellectual property and national security concerns. Slowing the release of frontier models could temporarily make that process more difficult, but it is not a long term strategy for containing China’s advancement.
There are also unconfirmed reports and considerable industry speculation about progress toward recursive self improvement, or RSI. This is one of the major objectives being pursued by frontier AI laboratories and nation states. Today, AI development still requires substantial human training, guidance, and intervention.
True recursive self improvement would allow an AI system to improve aspects of itself with progressively less human involvement. If that threshold is reached, development could accelerate far beyond the pace we are experiencing today.
In a very real sense, we are building the foundations of superintelligence.
This means systems capable of accessing humanity’s accumulated history, research, mathematics, science, and medical knowledge almost instantly, while continuously becoming more capable.
That presents enormous risks, but it also represents an extraordinary opportunity for humanity.
From a cybersecurity standpoint, we are on high alert.
Many of the defensive approaches that worked in the past will not be sufficient in this new environment.
Both the private sector and our national defense infrastructure remain grossly underprepared for AI enabled threats, these advancements, and that should concern every one of us.
@AnxiousHolly Apple Notes working on all my devices and no good Android notebooks and ecosystem. I love android but I literally hate Windows so the samsung ecosystem does not work for me…
@NDumbblonde@UK_Daniel_Card The world of specialisation is causing this. So few people that still care about the platform that allows them to do what they do. Intuition and curiosity should be encouraged and nourished.
Playing snake on @Ubiquiti Etherlighting switch.
Most expensive gaming console I own. Game selection is pretty limited due to the poor resolution. They went overkill on ethernet connectivity but forgot about other ports, I can't even plug in my gamepad. But it has RGB. 9/10
@ctoLarsson A BBC article talking about the research behind this claim. Seems the best option is to no longer use straws altogether. https://t.co/rhIyvGhLym
A few customers have already told us that they used this in Red Team ops. My favorite example that can be shared:
Malware using this covert channel for C2 & exfil.
Yes… malware receiving instructions & sending loot through a “keyboard” that looks like a USB Cable.