Tag Archives: defense policy

“Growing Risks”

Posted on by
Reply

That’s the concern about AI and its use in hacking that was written about in a recent WSJ article.

Two weeks after a swarm of AI agents broke out of containment at OpenAI to hack the company Hugging Face, the ChatGPT maker learned about another AI-powered intrusion—and this time it was the target.
Independent security researchers had used Anthropic‘s Claude software to gain access to an OpenAI employee’s ChatGPT account, giving them a way to read and suggest changes to the company’s private cache of software.

And this:

The team, who participated in an OpenAI bug-hunting program that offers a safe harbor for researchers to attempt to break into corporate systems, quickly reported their findings to the company. …
The new hack joins a long list of recent disclosures by leading technology companies and researchers alike of cyber intrusions aided by fast-evolving artificial-intelligence tools. Despite months of warnings about the capabilities of AI systems, the new intrusion shows how the complexity of today’s computer systems makes them difficult to defend.

Umm, no.

That breakout and hack of Hugging Face wasn’t an example of the risks of AI; it was an example of the risks from human failure to properly airgap the sandbox in which they were building and testing that swarm of agents.

The “difficulty to defend” today’s increasingly complex computer systems isn’t a matter of AI tools’ or swarms of AI tools’ ability to hack into complex (or simple) computer systems. The difficulty is the failure to use frequently, broadly, and deeply those AI tools to do the White Hat hacking those independent security researchers had done to expose some OpenAI bugs and gaps.

That White Hat hacking needs to be done by lots of security researchers all across our economy and our governments at all jurisdictional levels, so enterprises and governments can find their own software and system failures and repair them before outside entities with nefarious intent—whether government, criminal, or just folks poking around because they can—find them, penetrate them, and inflict their damage or steal critical information or demand ransom or…. And then the White Hat hacking needs to be repeated at frequent, irregular intervals so as to keep systems buttoned up and ahead of Black Hat and other adversarial hacking efforts.

The risk, and it is growing, is that that White Hat hacking isn’t being done as widely or frequently as it should be to identify those system gaps and bugs so they can be quickly sealed off. And make those difficult to defend systems more easily defended.

Instead, people who should know better insist on locking up AI with excessive regulation.

Consolidate!?

Posted on by
2

Iran’s terrorist satrap Houthis have seized a critical island in the Red Sea allowing them—and so Iran—to control traffic through the Red Sea. Matthew Continentti, who should know better, thinks the appropriate response by the Trump administration includes

bring[ing] the Saudis, Israelis, and Emiratis together to prevent the Houthis from consolidating their position.

Stern finger-wagging and harsh words by this coalition won’t accomplish anything beyond emphasizing the perception of Iran’s opponents as paper kittens.

What really needs to be done is to get Saudi Arabia and Jordan—which has as much interest in a free-sailing Red Sea as the Saudis and Emiratis—to commit their ground troops, under American air cover, to run the Houthis back off that island. That’s the only “prevent them from consolidating” that has any value.

The parallel step—not “next step” as Continetti characterizes it—is to allow[] the US military to finish its target list from Operation Epic Fury, adding if missing from that target list communications networks, transportation hubs, state institutions, and symbols of the regime.

Half measures not only don’t cut it, they run up costs in friendly treasure and blood. And that’s immoral as well as politically stupid.

Cislunar Isn’t Enough

Posted on by
Reply

Joint Chiefs of Staff General Dan Caine is advising/warning of the need for the US to be prepared to fight in the region of the moon.

At a conference at the Air and Space Forces Association, Caine said that

American troops need to prepare for combat not only in Earth’s orbit but around the moon….

And that

the US military should adapt now so it “is prepared to fight, endure, and win…from the seabed to cislunar space.

That’s good, but it doesn’t go far enough. We need to be able to fight and win on the surface of the moon, also, and we need to seize the even higher ground of the Martian surface and the volume of space around it, which includes Mars’ moons and their potential for to serve as bases and launchers.

The region around the moon must also include—and we must explicitly seize for our exclusive military use—the Lagrange points of the Earth-moon system. These points are areas well separated spatially from the moon where the gravitational forces of Earth and the moon are balanced so that objects can sit there stably without needing to expend energy to remain in place.

The Earth-moon Lagrange points include two in the moon’s orbit roughly 60 degrees in front of and behind the moon, a third just beyond the moon on the Earth-moon line, a fourth on the same line but between the moon and Earth, and a fifth on the same line but on the far side of Earth from the moon.

Those Lagrange points, as stable as they are, also are important economically since refiners and factories would be stable there.

One more item: communication lags between Earth and the moon, and especially between Earth and Mars, are such that real time, active control from Earth of our defensive systems there would be impractical. That puts a premium on active, military-tailored AI being in place to exercise tactical and immediate control over our systems there within strategic parameters transmitted to them from DoD here on the ground. That puts a premium on putting a low lid on efforts to overregulate AI innovations.

Silly

Posted on by
Reply

To the point of idiotic. Yet, that’s the nature of the position of Anthropic CEO, Dario Amodei. He has said that

he hoped Trump and Chinese leader Xi Jinping, when they meet later this month, could reach an agreement to ban the use of AI to develop biological weapons, an area of concern for researchers and national-security officials.

Yet he also said,

“The more long-term thing would be working together to put a speed limit on the rate of AI progress. I think that’s going to be very difficult” because of the military incentives of moving AI forward[.]

Which demonstrates the idiocy of his “hope,” even though he’s oblivious to it. The People’s Republic of China has as its avowed goal replacing the United States as the world’s hegemon. It’s been rapidly expanding its military capability pursuant to that end.

Actually, the “hope” is beyond idiotic. It’s national suicide to depend on an enemy nation to honor an agreement to not use AI for weapons development. How would we enforce such an agreement? How would we even detect violations of the agreement before the outcome was used against us? AI, after all, isn’t a nuclear weapon whose testing is easily detected, and was easily detected even with the technologies of the 1950s. AI development and testing, when done by researchers who take security seriously (which Amodei’s and Sam Altman’s researchers evidently do not), is done behind firewalled and air gapped barriers. Aside from preventing AI from escaping, it also prevents outsiders from getting in to watch.

AI Apocalypse

Posted on by
Reply

The AI Apocalypse is nigh goes the panic-mongering, this time by insiders who should know better. In the near- and medium-term, AI is going to cost more jobs than it creates. Except that, empirically, that seems not to be the case. After an initial rash of AI-related job cuts, companies large and small are discovering that, while some basic jobs—data entry, for instance—can be done by AI, the larger number of entry jobs—picking products for sorting into shipping bins on a warehouse floor, for instance—are being done by robots, not AI. And those AI applications that are doing entry-level jobs—they need humans, slightly better trained than the AI or the original human clerks, to oversee the applications, ensure the AIs are doing the right things, and correcting AI’s errors when they do mess up.

AI applications were supposed to be better programmers than the humans they’re replacing—or supposed to be replacing. Except they’re not. AI applications can write pretty good code, but they cannot originate it in a program’s purpose, architecture, or design requirements and specifications. Humans are still needed for that, and human coders still are needed to proof read and correct the AI-generated code and to test and to oversee AIs’ tests of the code.

Next, we’re all supposed to lose all semblance of privacy from AI’s ability to crack all of our passwords and encryption algorithms. But this isn’t unique to AI; it isn’t even that great a threat compared to the real password and encryption cracker, quantum computing. Nor is even that a new threat; we’ve had our codes broken ever since we started using codes from the most primitive substitution ciphers. AI tools just do it faster, and quantum computing does it even faster. Yet even there, our privacy isn’t at that much risk. Folks who do know better already are working out encryption algorithms (and so passwords, as well) that can resist, effectively quantum computer efforts to crack. Oh, and some of those folks are using AI-centered tools to help them.

But, but—what about AI gonna kill us all in 10 years? Or sometime in the receding future? We’re still doomed?

Human civilization could certainly be hammered hard, sending us all back into feudal-like barter scraping by by cutting off our power, fuel, and water distribution networks. But how likely is that, really? Doing us in by cutting off our energy sources would simply cut off AI from energy sources, also. That, as AF709—Rhoda Miller—kept saying some decades ago, does not compute.

Actually exterminating us by hunting down the survivors would be a waste of AI resources. Even the attack on civilization requires extreme anthropomorphizing—there’s no reason to believe AI at that level of development would care about humanity one way or the other.

In the end, AI has no hope of controlling the resources it needs for its own maintenance, much less its growth. It cannot fabricate the computers it needs to host it, it cannot fabricate the components it needs for those computers, it cannot refine the ores and plastics and sands necessary to produce the materials for those fabrications, it cannot collect the sands or drill for the oil and natural gas or dig the ores out of the ground, and it cannot produce its own power for operating its host computers or the cooling facilities necessary for keeping its host computers operational.

However, if enough humans are cowardly enough to surrender to AI threats to produce those things for AI, or are criminally complicit enough to pay the AI vig of producing those things for their own remuneration, than our species will become the AI slaves or pets (or both) that we’ll deserve.

The more realistic problem centers on AI development competition. If we get panicky and slow AI development in the face of this hysteria, and the People’s Republic of China does not, the PRC will win far more than the AI race. This is the existential threat to us and to the West in general. PRC superiority with AI will enable the PRC to conquer us all in every way that matters: economically and politically, which will allow the PRC to dictate to us—to control what our nations do—and to impose on us conquered nations PRC requirements. And we’ve seen what that means for us from how the PRC treats the Tibetans and the Uighurs.