This is the script of my national radio report yesterday on the very bad week for AI. As always there may have been minor wording variations from this script as I presented this report live on air.
- - -
Yeah, this has been a remarkably bizarre week for AI, oh but before we get into that I do have a bulletin. Remember we've discussed in the past about how the U.S. FCC was talking about banning consumer grade Internet routers made outside the U.S. just as they did for drones, and on the same basis. They claim they have "security concerns" but have never described or demonstrated them publicly, so it's hard not to feel like they're treating us as ignorant fools with these claims. Anyway, the FCC went ahead and did ban consumer grade Internet routers made outside the U.S. And this of course affects a far larger number of people than drones, because if you've got Internet in your house, use Wi-Fi and so on, you probably have a covered router, and as a practical matter none you're likely to have are built in the U.S. So it's going to be a mess. Existing models can still be manufactured, imported and sold, but as new models are designed with better performance and even better security, U.S. consumers are blocked from them under the current rules unless exceptions are made. I'll discuss this more in the future.
OK, about AI's awful week. So some days ago, Anthropic made a little boo boo. Well, actually, a very big boo boo. They accidentally publicly posted the source code for their Claude Large Language Model AI on GitHub, a very popular source code repository. They quickly realized their error and closed access, but by then it had ALREADY been noticed and copies were propagating everywhere. Anthropic reportedly got thousands of copies pulled down but it's still out there and even already translated into other computer languages.
And while this isn't going to crush #Anthropic or anything like that it did give quite a bit of insight into Claude's design and likely future features including agentic capabilities and lots of other things.
Perhaps even more relevant is that this kind of screw up doesn't exactly give anyone much faith in how safe Anthropic's operations are overall and that just adds again to the mass of serious concerns about Generative AI and how Big Tech is managing it.
BUT WAIT THERE'S MORE!
You may have heard of OpenClaw -- and no this isn't a character from an old TV series like Get Smart or Inspector Gadget. #OpenClaw is a public open source agentic #AI that very quickly became enormously popular because it could be so easily deployed. And that enthusiasm among AI boosters wasn't really held back by concerns that all sorts of possible problems and exploits might be part and parcel of OpenClaw, and now sure enough that's the case. Without getting into technical details it's a pretty bad situation and frankly as you can imagine hasn't come as a surprise to many observers.
If you're somewhat, uh, shall we say, skeptical of? concerned about? these AI systems, as obviously I and many others inside and outside the technical community definitely are, these events inspire even greater concerns about these systems.
Because now we're seeing more evidence of the inherent problems, not just as deployed by Big Tech but even as public open source systems. This isn't just a matter of big corporations trying to exploit us by shoving this tech down our throats, but also a set of fundamental problems that are likely inextricably linked to the very concepts involved with the creation and deployment of these Large Language Model AI systems, whether commercial or free, closed source or open source. And agentic AI, which is where the big push is now, is of peak concern, since running amok they can do enormous damage to users.
These kinds of events cannot be minimized or ignored. They are flashing red warnings of even more numerous and serious problems yet to come. And if we don't take these warnings as a clear signal for meaningful regulation and control, we'll have nobody but ourselves to blame when future AI fiascos make today's AI problems look like vacations in comparison.
- - -
L
