It was a disastrous week for cryptocurrency in the United States, as the Securities Exchange Commission (SEC) filed suit against the two biggest exchanges, Binance and Coinbase, on a theory that makes it nearly impossible to run a cryptocurrency exchange that is competitive with overseas exchanges. Nick Weaver lays out the differences between “process crimes” and “crime crimes,” and how they help distinguish the two lawsuits. The SEC action marks the end of an uneasy truce, but not the end of the debate. Both exchanges have the funds for a hundred-million-dollar defense and lobbying campaign. So you can expect to hear more about this issue for years (and years) to come. I touch on two AI regulation stories. First, I found Mark Andreessen’s post trying to head off AI regulation pretty persuasive until the end, where he said that the risk of bad people using AI for bad things can be addressed by using AI to stop them. Sorry, Mark, it doesn’t work that way. We aren’t stopping the crimes that modern encryption makes possible by throwing more crypto at the culprits.  My nominee for the AI Regulation Hall of Fame, though, goes to Japan, which has decided to address the phony issue of AI copyright infringement by declaring that it’s a phony issue and there’ll be no copyright liability for their AI industry when they train models on copyrighted content. This is the right answer, but it’s also a brilliant way of borrowing and subverting the EU’s GDPR model (“We regulate the world, and help EU industry too”). If Japan applies this policy to models built and trained in Japan, it will give Japanese AI companies at least an arguable immunity from copyright claims  around the world. Companies will flock to Japan to train their models and build their datasets in relative regulatory certainty. The rest of the world can follow suit or watch their industries set up shop in Japan. It helps, of course, that copyright claims against AI are mostly rent-seeking by Big Content, but this has to be the smartest piece of international AI regulation any jurisdiction has come up with so far. Kurt Sanger, just back from a NATO cyber conference in Estonia, explains why military cyber defenders are stressing their need for access to the private networks they’ll be defending. Whether they’ll get it, we agree, is another kettle of fish entirely. David Kris turns to public-private cooperation issues in another context. The Cyberspace Solarium Commission has another report out. It calls on the government to refresh and rethink the aging orders that regulate how the government deals with the private sector on cyber matters. Kurt and I consider whether Russia is committing war crimes by DDOSing emergency services in Ukraine at the same time as its bombing of Ukrainian cities. We agree that the evidence isn’t there yet.  Nick and I dig into two recent exploits that stand out from the crowd. It turns out that Barracuda’s security appliance has been so badly compromised that the only remedial measure involve a woodchipper. Nick is confident that the tradecraft here suggests a nation-state attacker. I wonder if it’s also a way to move Barracuda’s customers to the cloud.  The other compromise is an attack on MOVEit Transfer. The attack on the secure file transfer system has allowed ransomware gang Clop to download so much proprietary data that they have resorted to telling their victims to self-identify and pay the ransom rather than wait for Clop to figure out who they’ve pawned. Kurt, David, and I talk about the White House effort to sell section 702 of FISA for its cybersecurity value and my effort, with Michael Ellis, to sell 702 (packaged with intelligence reform) to a conservative caucus that is newly skeptical of the intelligence community. David finds himself uncomfortably close to endorsing our efforts. Finally, in quick updates: Nick talks about Tesla’s Full Self Driving, and the acci