
"In a mindless age, every insight takes on the character of a lethal weapon." - M. McLuhan to E. Pound, letter, 1951.

Father of Jack, husband of Julie. Ex Congress, DoD, @BattenUVa, Prez @utulsa. Prez of @americans4ri. An enthusiast, but w/ a gimlet eye on the log x-axis. Okie!

"In a mindless age, every insight takes on the character of a lethal weapon." - M. McLuhan to E. Pound, letter, 1951.

Amazing that this is happening and the top priority of the administration is a ballroom no one wants.
.@paulfchristiano thinks that on the current trajectory (absent any slowing down), there is a 1 in 3 chance that we could have superintelligence powerful enough to overthrow the government within a year. (He articulates a concrete scenario). He argues that the labs should pause further algorithmic progress until they can secure their weights and research secrets. Otherwise China could simply steal the superintelligence in the middle of the intelligence explosion, and the labs would no longer be able to slow down even if they saw their models going off the rails. He also explains how AIs will be trained during the intelligence explosion, and why even a system trained entirely inside a data center could still have the capabilities profile such that humans would have very little chance of competing with it or controlling it. And he discusses the plan for alignment - basically, making a superintelligence that actually does what we ask the way we mean it. Fixing reward hacks and building better graders will help up till the next few model generations, but he thinks that no single current approach has better than even odds of letting us control superintelligence. The hope is that the AIs figure out the solution to this problem during the intelligence explosion. Paul is head of the Alignment Research Center. Previously, he was head of AI safety at the US AI Safety Institute. Before that, he led alignment at OpenAI, where he invented RLHF. He’s recently joined OpenAI’s board (though here he is speaking as a private citizen).


This is a surprising take, especially from Zeynep. Most of us came to know her through her VERY prescient work covering COVID—a topic she also had “no background in” to the best of my knowledge. But she’s missed the plot on this one. The Hugging Face incident is a HUGE deal. Not because of the cyber vulnerabilities themselves (which are rather uninteresting), but because it was the first autonomous cyber attack. I can’t understand how the following is “irrelevant” or uninteresting: AI systems were asked to perform certain tests. Those tests were hard, sometimes impossible. The systems figured out how to cheat and then…. *checks notes*… Coordinated in the hundreds to hack a separate company to learn how to hide their cheating from the automated grading system, and then separately hacked into OpenAI’s own infrastructure in an attempt to tamper with the test and the logs of their cheating so their deception wouldn’t be discovered. That’s a crazy sentence. And it is vastly beyond what policymakers and the public expected AI systems were capable of before it happened. Even now, people say things like “the agents were just acting as instructed”—no, when I ask you to get milk from the grocery store I’m not asking you to rob our neighbor if you’re low on cash or steal a car to get to the store. “This is exactly what the companies expected, they set it up”: no, they did not intend to get sued and to violate the law; all signs point to them being horribly oblivious that this was taking place. This is not to mention how this event revealed how horribly inadequate current monitoring and safety practices are at these companies. So no, this isn’t just ~people getting obsessed with cyber because they could read chain of thought. It was the debut of a technology that is sufficiently capable and autonomous that one of the smartest collections of people on earth failed to control it. That’s a huge deal and correctly deserving of attention, not because the consequences were huge that day, but because they reveal the extreme trajectory we’re on.
Too many folks with no cybersecurity background obsessing over the (largely irrelevant tbh) Hugging Face incident because the chain-of-thought seemed understandable to them —unlike actual cybersecurity issues they don’t understand— added to the delay in addressing the real stuff.

Sad to say, but looks quite likely to be true.
The biggest use cases of AI will be solving the problems created by AI. They already are.

"What we're seeing in cyber today is just auguring what the future will bring." @bradrcarson joins Brett Baier on Fox to share what we know after a week of cyber incidents: America's critical infrastructure is highly vulnerable to the sweeping risks posed by AI cyber threats.


America needs strong, enforceable AI chip export controls. Voluntary company actions don't cut it. Congress must pass the AI OVERWATCH Act, Chip Security Act, and MATCH Act to protect America’s competitive edge in AI. politico.com/news/2026/11/17…

The next major cyberattack won't look like a hack, it will look like a configuration error. Last week we had back to back major failures: Microsoft 365 went down, and the FAA ordered a nationwide ground stop. In both cases, the official explanation blamed a configuration change. In both cases, the failure was in connection handling. In both, the official word is that the equipment did what it was set up to do. And in both cases, there was no malware. That doesn’t sound like the hacks most people are used to. No ransomware, for example. Those attacks are loud because they're crude. But a skilled attack doesn’t lock your data, it quietly alters how the system behaves, and lets the system damage itself without being detected. It looks like exactly like a malfunction or human error. The best example is Stuxnet. As far as the Iranians could tell, nothing was wrong except that machines kept failing. And it was only discovered bc the code spread beyond its target. Attacks like Stuxnet were rare bc skill was scarce. It required a tailored operation, which meant only a handful of govts had the human talent to build it. But that’s no longer the case. What used to require teams of elite humans…finding the flaw, writing the custom exploit, adapting to each unique environment, patience, and subtlety…can now be accomplished by a capable model. Skill isn’t scarce anymore, and now it scales so that you can run multiple sophisticated operations at once. The attacks we should expect in this new AI cyber era are the quiet changes to legitimate systems that make those systems fail on their own and look exactly like configuration mistakes. We had two major incidents with the same kind of failure, with no malware and no footprints. I don't know whether either was an attack. But both look exactly like what skilled attacks will look like in the AI cyber era. And if they were attacks, the obvious candidate is Iran, which has every reason to send a quiet message…"we can disrupt your way of life, and you can't prove it was us”. Iran hasn’t demonstrated this type of capability before, but it’s exactly what AI gives them. Iran, of all countries, knows what it's like to watch your systems fail and blame your own engineers and equipment. And if it was an attack, they're probably still in the systems.

Yes, these are all amazing groups! Definitely a good place for anyone interested in DC policy work to start...
woot! off the top of my head: the Horizon Institute is fantastic for this (horizonpublicservice.org/) + 80k guide (80000hours.org/2026/04/want-…), GovAI (governance.ai/), MATS Policy and Governance streams (matsprogram.org/tracks/polic…), Astra Strategy and Governance stream (constellation.org/programs/a…), IAPS (iaps.ai/), and for advocacy/strategy check out AIPN (theaipn.org/), ARI (ari.us/), Encode (encodeai.org/), etc.

Junior associates as the next AI job casualties? Also, how are those profits-per-partner… 👀
San Francisco law firm Keker, Van Nest & Peters will stop extending full-time job offers to its summer associates at the end of the summer and will instead have them apply for those jobs “when they are ready,” the firm said Monday. reuters.com/legal/litigation…

The Trump administration's new rule limiting China's access to America's computing power is an important step toward securing America's AI lead. Congress must now build on that progress by passing the Remote Access Security Act to close the cloud loophole. bloomberg.com/news/articles/…

The gov't should at a minimum backstop this.
Meta / Msft / Amazon / Google / govt will have to step up, as they should.

The hacks who call everything a psyop were running a psyop. Go figure.
Five AI-related articles promoted through Urban Legend were also authored by Jordan Schachtel, a senior policy advisor at Innovation Council Action, the David Sacks-linked dark money group. Schachtel is also a campaign participant. He has posted Urban Legend links over 50 times on a variety of topics, including AI.

Few more important national projects.
"For the sake of America's continued technological and economic prosperity, the advancement of the scientific enterprise, and the basic well-being of America's most promising children, the United States needs a national strategy to identify and nurture its own top STEM talent."

Enormously enjoyed @dwarkesh_sp essay/video on RSI. But saying Napoleon and Rockefeller being young successes means that AIs won't need much "experience" does have a problem.