Bitcoin slid below 63,000 dollars on Friday as risk assets sold off across the board, with miners squeezed and bearish options bets piling up all the way down to 52,000. Strategy added another 1,587 BTC for 100 million dollars, lifting its stack to 846,842 coins. Anthropic had its two most powerful models, Fable 5 and Mythos 5, yanked offline by a U.S. government export-control order. And in legal tech, Harvey announced it's open-sourcing custom legal models while Italian law firm Legance rolled out Legora across every practice area. Here's what actually matters.
Bitcoin is having a rough week. Price dropped below 63,000 dollars on Friday, the fourth straight down day, giving back the entire bounce in holiday-thinned trading. Oil fell 9% as ships moved through the Strait of Hormuz for the first time in weeks, the Iran deal got signed, and somehow the calmer geopolitics translated into crypto selling rather than buying. Capital is rotating into AI stocks instead.
The real story underneath is the miner squeeze. Bitcoin has traded below the average cost of production for five straight months. About 20% of miners are unprofitable right now. Publicly traded miners dumped more than 32,000 BTC in the first quarter alone, more than they sold in all of 2025. That's structural sell pressure that doesn't care about your chart patterns.
Then there's the STRC situation. Strategy's variable-rate preferred stock, paying around 11.5%, has been struggling to hold par value. Strive CEO Matt Cole blamed forced selling from leveraged investors for the digital credit market selloff this week. Saylor's response was to shift STRC dividends to twice-monthly payments to stabilize things. Meanwhile Strategy keeps buying, 1,587 BTC last week funded by selling 209 million dollars of MSTR shares through the at-the-market program. USD reserves are up to 1.1 billion. The total cost basis is now around 64 billion against holdings worth roughly 56 billion at current prices, so they're sitting on about 8 billion in unrealized losses.
Bitwise analysts say Bitcoin is trading at a historical discount to AI stocks, which sounds bullish until you remember hawkish Fed signals are competing for the same liquidity. One trader is calling for a Q3 macro bottom near 50,000 dollars before any real reversal. The bear bets piling up at 52,000 suggest options markets are taking that seriously.
This one is wild. Last Friday afternoon the U.S. government handed Anthropic an export-control order forcing them to immediately disable Claude Fable 5 and Claude Mythos 5 for every user worldwide, including their own employees abroad. Anthropic complied, then publicly disputed the decision.
The government's stated reason is a jailbreak of Fable 5. Anthropic's counter is that the jailbreak isn't universal, it's just prompting the model to read a codebase and identify flaws, and that exact capability exists in GPT-5.5 and is used routinely by cybersecurity professionals for defense. Mythos 5 was barely deployed anyway, restricted to about 50 vetted organizations under a controlled program called Project Glasswing because of its vulnerability-discovery abilities.
Here's the irony. Anthropic built its entire brand on safety-first AI, on warning loudly about what frontier models could do. Sam Altman called Mythos marketing fear-based. And now the government has taken Anthropic at its word and pulled the plug. The safety warnings became the export-control rationale. Anthropic is heading toward an IPO, and their most capable products just got shut off by Washington over what they argue is a narrow issue. If this precedent holds, any new frontier model that demonstrates a jailbreak could be killed on arrival.
Meanwhile, on the product side, Anthropic shipped Claude Code Artifacts for Team and Enterprise. These are live, shared HTML dashboards that update in real time as Claude runs, plug into local repos and monitoring tools, and stay inside corporate authentication boundaries. They also overhauled Claude Design with design-system imports, round-trip integration with Claude Code, and export to Figma alternatives like Canva, Replit, and Vercel. It's a clear platform play. Whether any of that matters if regulators keep yanking the underlying models is the open question.
Legal AI just had a genuinely interesting week. Three stories, one trajectory.
Harvey announced it's building custom legal foundation models and plans to open-source the models and training data. They're expanding Harvey Labs under an ex-Google Brain researcher, and the goal is agentic systems that can coordinate legal tech tools and frontier models like GPT-5 to handle complex matters that normally need a team of associates. They're also running proof-of-concept work with law firms to train open-source models on how those firms actually operate, including digital twins of how specific lawyers handle specific long-standing clients. The pitch is that data security plus targeted training plus agentic orchestration makes post-trained open-source models viable again.
Then Crosby, the AI-native law firm that's raised over 85 million from Sequoia, Index, Lux, and Bain, launched a contract negotiation benchmark called Redline. It tests how frontier models handle multi-turn negotiations, not just contract review. ChatGPT 5.5 leads at 50.5%, Claude Fable 5 hit 47.3% before access was cut, Gemini 3.5 Flash and Claude Opus 4.8 trail behind. The interesting finding: humans still beat AI at finding creative negotiation routes. The models tend to stay anchored on their initial positions. There's a judgment layer that hasn't been cracked yet.
And in Italy, Legance, one of the country's top independent firms, rolled out Legora across every practice area, every lawyer, every level, in three months. LinkSquares also went GA with an all-agentic contract lifecycle management platform.
The pattern is clear. The first wave was AI for document review and drafting. The second wave, which is happening right now, is AI that owns end-to-end workflows: intake, negotiation, signature, post-signature obligation tracking. Fixed-fee billing models become viable when an agentic system runs the matter and the lawyer is the judgment layer on top. Associates should pay attention. So should anyone billing hourly.
Botanix Labs is shutting down Spiderchain, its Bitcoin layer-2, after roughly one year on mainnet. They raised 11.5 million from Polychain and Placeholder, processed 25 million transactions across 200,000 wallets, and the economics still didn't work. Users have until July 9 to withdraw, after which a federation validator group takes over remaining funds. Counterparty risk for anyone who doesn't move in time.
The Botanix story is the honest one. Bitcoin-native DeFi demand just isn't there at the scale needed to cover infrastructure costs when BTC is chopping between 61 and 65 thousand and altcoin liquidity is drying up. That's the reality check for an entire category of projects pitched during the last cycle.
But not everything in Bitcoin L2 land is dying. Citrea launched mainnet as a Bitcoin ZK-rollup using BitVM and a Clementine bridge, with DEXes Satsuma, JuiceSwap, and Fibrous, lending via Morpho, and a Bitcoin-oriented stablecoin called ctUSD from MoonPay. Jameson Lopp called it the next major experiment in creating sustainable demand for Bitcoin block space, which is a careful, accurate framing. OP_NET launched a different approach, smart contracts running directly inside standard Bitcoin transactions, BTC as the only gas asset, no bridges, no wrapped BTC. They're calling it SlowFi, which is at least an honest name.
On the payments side, GoMining released an SDK for GoBTC Pay, going directly at Jack Dorsey's Square with a merchant payment system designed around Bitcoin from the ground up. And Bitcoin network activity is near record highs thanks to a microtransaction surge driven by OP_RETURN usage, even with the price grinding sideways.
The takeaway: the Bitcoin L2 thesis isn't dead, but the version that copy-pasted Ethereum DeFi onto Bitcoin and expected the same flywheel is getting buried. What survives looks more like payments rails, ZK settlement, and protocols that treat Bitcoin's block space as the actual product.
Quick prediction: by the end of this cycle, the most valuable Bitcoin L2 won't be the one with the most TVL. It'll be the one that processes the most payments nobody notices they're making.