Anthropic just put a Mythos-class model in public hands for the first time, and a researcher claims he already jailbroke it. Bitcoin is clinging to $63,000 with miner margins at record lows and corporate buying near zero. Metaplanet's stock is trading below the value of its bitcoin, and the CEO is openly talking buybacks. Botanix, a four-year-old Bitcoin layer-2, is shutting down and telling users to withdraw before July 9. Japan just passed a sweeping bill to regulate crypto like stocks. And Digital Asset raised $355 million to bring Wall Street onchain. Let's get into it.
Anthropic released Claude Fable 5 on June 9, and this one matters. It's the first time a Mythos-class model has been made available to the public. Mythos is Anthropic's internal designation for what they call a step change in capability — the tier they previously held back from general release because of cybersecurity concerns.
The rollout is split. Fable 5 is the public version, wrapped in safeguards. Mythos 5, the unrestricted sibling, goes to vetted partners through something called Project Glasswing — cyberdefenders, critical infrastructure operators, and select government-adjacent groups. Same underlying model, different leash.
The safety architecture is interesting. Fable 5 runs classifiers on incoming prompts. If something trips the cybersecurity or biology filters, the query gets quietly rerouted to the older Opus 4.8 model. Anthropic says this happens in under 5% of sessions, and they claim external red teamers spent over 1,000 hours trying to find a universal jailbreak and failed.
That claim aged about 36 hours. A researcher who goes by Pliny the Liberator says he's already found holes in the fence. Whether that's a true bypass or a narrow trick, the layered safety pitch is going to get tested hard in public.
On capability — Anthropic says Fable 5 leads on 11 of 13 benchmarks, with strong vision, chart and diagram extraction, and code reconstruction from screenshots. It's built for multi-day autonomous work: planning across stages, delegating to sub-agents, verifying its own output. Pricing is $10 per million input tokens and $50 per million output, with a 90% discount on cached inputs. Free through June 22 for Pro, Max, Team, and Enterprise subscribers, then it shifts to usage credits.
The bigger story here is the two-tier distribution model. Public model with guardrails, restricted model for trusted parties. That's a real shift in how frontier AI gets deployed — and probably a preview of how every lab handles the next tier up.
Bitcoin is hovering around $63,000, and the structural picture under the price is ugly.
Start with demand. ETF flows have turned negative, and the narrative that investors are dumping bitcoin to load up on the SpaceX IPO doesn't quite hold up. Sygnum's Fabian Dori argues this is mostly arbitrage unwinds — basis trades closing out as funding rates flipped — not retail FOMO into Elon's rocket company. Either way, the bid is gone.
Corporate buying has fallen off a cliff. At the peak, treasury companies were absorbing roughly $500 million of bitcoin per day. That number is now near zero. The two demand pillars that defined the last cycle — spot ETFs and corporate treasuries — are both quiet at the same time.
Glassnode says over 95% of short-term holders are underwater, and realized losses are approaching capitulation territory. That sounds like a bottom signal, except the same analyst warns the hard part is the slow grind that follows. Their view: a durable recovery probably needs the dollar index to break below 99, or the 10-year Treasury yield to compress meaningfully. Neither is happening right now — US inflation just printed a 3-year high, and the ECB hiked for the first time in nearly three years.
Miners are getting squeezed too. Margins hit a record low this week. And there's a Bank of Japan rate decision pending — historically, BOJ hikes have produced an average 22.5% bitcoin selloff.
Meanwhile the ETF business is consolidating. BlackRock's IBIT and Fidelity's FBTC are now capturing essentially all the net inflows. Smaller funds are being squeezed out. BlackRock is also about to launch an income-paying bitcoin ETF that sells covered calls on its own IBIT, undercutting rivals on fees.
So: weak demand, capitulating holders, miner stress, macro headwinds, and the ETF market collapsing into a duopoly. The setup isn't broken — but it's not the setup people were buying six months ago.
Metaplanet is the cleanest stress test we have right now of the bitcoin treasury company model in a drawdown.
The Tokyo-listed firm holds 40,177 bitcoin. That makes them Asia's largest corporate holder and number three globally. Bitcoin NAV is around $2.54 billion. Enterprise value is around $2.35 billion. Which means the market is pricing the company at less than the bitcoin on its balance sheet. mNAV — market cap divided by bitcoin NAV — has dropped to 0.92, briefly hitting 0.90.
When the premium flips to a discount, the playbook inverts. You don't issue equity to buy more bitcoin — that destroys bitcoin-per-share. Instead, you buy back stock. CEO Simon Gerovich confirmed this week that with mNAV under 1.0, buybacks are firmly on the table, and the lower the mNAV, the more accretive they get. The company already has a $500 million repurchase program authorized.
The stock tells the story. Metaplanet shares are down about 90% from the 52-week high, down 44% year-to-date, down 35% just in the last month. Unrealized losses on the bitcoin position are around $1.64 billion.
There's also a structural maneuver being floated. Metaplanet wants to issue a US-style perpetual preferred stock — something like Saylor's STRC at Strategy — but Japanese market rules make a monthly-adjustable preferred hard to execute quickly. One theory making the rounds: Metaplanet could acquire Nakamoto, the struggling US-listed treasury vehicle, to inherit a Nasdaq listing and run the preferred through that. Nothing's confirmed. It's a fit on paper with non-trivial regulatory and trust hurdles.
And Strategy itself — the original of the species — is back in the spotlight. Michael Saylor and Jack Mallers had a public spat this week over Strategy's reporting metrics, mNAV, and whether issuing equity for cash dilutes or strengthens shareholders. Saylor's argument: as long as you're buying bitcoin below intrinsic value per share, issuance is accretive. Mallers isn't buying it.
The treasury company thesis was never tested in a sustained drawdown. It's being tested now.
Botanix is shutting down. The tokenless Bitcoin layer-2, EVM-compatible, four years in the making, is telling users to withdraw all funds by July 9. After that, the validator federation sweeps whatever's left and it's gone.
The numbers are not embarrassing. Botanix processed about 25 million transactions across 200,000 wallets. They integrated Chainlink and OKX. They launched BINK, a self-custodial bitcoin neobank. What they couldn't do was generate enough fee income to cover infrastructure costs, and without a token, they had no way to subsidize the gap.
The post-mortem is more interesting than the shutdown. The team's read on the market is blunt: most bitcoin holders treat their bitcoin as a store of value. They want yield, not high-frequency activity. And when they do want activity, they go to centralized venues — exchanges, Hyperliquid, and increasingly traditional finance rails — not decentralized layer-2s. Even users who claim to value decentralization end up on wrapped BTC on other chains because it's cheaper and easier.
That's a hard message for the entire Bitcoin layer-2 sector. If a technically capable team with real integrations can't make a tokenless model work, the implication is that base-layer decentralized infrastructure is losing to venues that own the user relationship.
The counterpoint comes from a different design. Second just deployed Bark on Bitcoin mainnet — an Ark-based layer-2 that pools many users into a shared structure without individual payment channels. Users hold virtual UTXOs as pre-signed receipts and can pay over Ark, Lightning, or onchain from a single off-chain balance. The pitch is custodial-Lightning convenience without the custody. Bark is going directly after Wallet of Satoshi and Cash App, not the smart-contract layer-2 crowd.
The distinction matters. Botanix tried to be a programmable platform for Bitcoin DeFi. Bark is trying to be cheap, self-custodial payments. The second category may actually have product-market fit. The first is now an open question.
One more thing to watch on Bitcoin itself: BIP-110, the proposal to restrict non-financial data in transactions, is under 10,000 blocks from its activation deadline. This is the most serious governance fight on Bitcoin in years, and it's coming to a head soon.
If the most powerful AI model now ships in two versions — one for the public, one for the trusted few — then frontier capability has officially stopped being a product question and started being a policy one. The pricing tier is the easy part. The access tier is what to watch.