Weekly Top Stories - 06/12/26

Welcome to Galaxy Research's Weekly Top Stories. Subscribe to get this newsletter delivered to your inbox every Friday morning.

In this week's edition, Will Owen's covers Pump Fun's new application to pay anyone to do anything; Marc Hochstein details the CFTC's new prediction market rules; and Lucas Tcheyan covers Anthropic's new Mythos class model.

Got feedback on this newsletter? Email Email [email protected]. We’d love to hear from you.

Market Update

The total crypto market cap stands at $2.31tn, up 5.95% from last week (when it stood at $2.18tn). Bitcoin's network value is 4.53% of gold's market cap. Over the last seven days, BTC is up 1.84%, ETH is up 0.04%, and SOL is up 1.02%. Bitcoin dominance is 55.25%, down 88 basis points from last week.

Pay ANYONE to do ANYTHING (What Could Go Wrong?)

Pump.fun launched a marketplace that lets anyone pay anyone to do anything, and within days a stranger had a misspelled memecoin ticker tattooed across his forehead.

On June 4, Pump.fun launched GO (pump.fun/go), its first major product since the launchpad that made it the dominant force in Solana memecoins. The pitch is four words: “Pay ANYONE to do ANYTHING.” Mechanically it’s an escrow marketplace. Post a real-world task, lock SOL against it, and anyone in the world can complete it for the bounty.

The flow is simple. A creator connects a wallet, writes a task with a deliverable and a deadline, and funds it, with a minimum of ~$5. Once the bounty is live, the money is irrevocable. The creator can’t pull it until the bounty expires or a winner is paid. Submitters complete the task, upload proof (video, selfies, etc.) and Pump.fun reviews. Approved submissions pay out instantly from escrow; if nothing qualifies by expiry, the creator reclaims the funds after a dispute window.

Pump.fun is the sole arbiter. It alone approves, rejects, modifies, or cancels any bounty. Its decisions are final and non-appealable. Creators can recommend a winner, but the platform has the last word. Officially it blocks fraud, illegal tasks, and X terms-of-service violations, and claims to “prohibit self-harm and violence.” In practice, enforcement has been case-by-case.

The platform already has ~754 live bounties, 2,928 submissions, ~$440K sitting in escrow, and ~$114K paid out so far. The top listing dangles ~$52,127 to place a bet on Howl (online gambling app) from the summit of Everest. One man tattooed a memecoin ticker on his forehead (the bounty misspelled it “BOUTYWORK”) then minted a BOUTYWORK token using the tattoo selfie as the image and ended up earning more in creator fees than the bounty would have paid. Others: quit your job on livestream, say “thoughts on CHILLHOUSE” 100,000 times, make and give 100 jars of pineapple Kool-Aid to homeless people. The darkest listings, including suicide-related tasks, have been pulled.

None of this is new for Pump.fun. GO is the financialized version of the livestreaming feature it shipped in late 2024, which drew heavy criticism for rewarding explicit content and self-harm in exchange for token attention. Founder Alon conceded that moderation “hasn’t been great” and said the team had doubled its moderation staff.

The platform’s token hasn’t shared in the attention. $PUMP set a record low at ~$0.00135 on June 5, down roughly 84% from its peak, despite a $350m buyback that has run since mid-2025 without moving the price. A 10b token unlock (1% of the supply worth $14m) is set to take place June 12. The launchpad remains a cash machine, so GO arrives from a position of revenue strength and price weakness.

Our Take

The escrow primitive is clever. Trust-minimized coordination between strangers and money is a thing crypto does better than anything, and GO is a clean expression of it. Strip the mechanism away, though, and what’s left is a machine that pays people to do unhinged things on camera, with a smart contract as the unblinking judge.

This is Black Mirror’s “Common People” rendered onchain. In the episode, a man takes requests from anonymous viewers and performs increasingly degrading, self-harming stunts on a livestream to cover his wife’s spiraling medical bills. Crypto Twitter flagged the parallel when GO launched, and it runs the same loop. Anonymous wallets, often flush from a memecoin trade, dangle life-changing amounts of SOL at strangers to humiliate themselves on camera. This is pure incentive alignment between money and spectacle.

It’s also the most public face of crypto for everyone who doesn’t trade it. Non-crypto people will see forehead tattoos and degens treating real life like a video game with respawns. GO crystallizes the exact caricature the industry’s critics reach for, and hands them the clip reel to prove it.

The problem is the design. “ANYTHING” is the product, and discretionary, after-the-fact review by the same company that profits from the volume is not a guardrail. Pump.fun has been clear about who decides which submissions get paid and vague about which bounties are allowed to exist in the first place.

GO will work, at least as marketing. Most viral stunts are free distribution for a token, and every bounty, lock and payout is SOL moving through the system. GO will drive numbers while handing crypto critics a permanent highlight reel. Pump.fun is betting, again, that attention is worth the backlash. So far, it’s usually been right.

The open question is whether “neutral referee” survives contact with the first listing that gets someone seriously hurt. Let’s hope no one does. - Will Owens

CFTC Proposes Prediction Market Rules

The Commodity Futures Trading Commission on Wednesday proposed new regulations for prediction markets that could meaningfully reduce uncertainty for platforms seeking to list novel event contracts.

The 267-page document clarifies the definitions of “involves” and “gaming,” seemingly straightforward words but whose meaning under federal commodities law was litigated during the previous administration’s court battle with prediction market powerhouse Kalshi. It provides a non-exhaustive list of contract types that fall outside the scope of public interest review by the CFTC and therefore would be easier to list (including economic indicators, forex rates, election results, legislative and appointment outcomes, and award contests). The proposal would also replace the existing, less-developed approach for public-interest review with a structured, multi-factor framework — asking, for instance, whether a contract provides meaningful hedging or price discovery utility, and whether it presents unusual risks of manipulation or insider exploitation. And the document spells out which sports-related contracts are unlikely to be found contrary to the public interest.

The proposal is open for public comment for 45 days after publication in The Federal Register.

Our Take

First, the big picture. This proposal comes at a time when prediction markets, once an academic hobbyhorse, have become a growth industry and a political football. More than 8,000 event contracts traded last month, compared to 220 for all of 2021, the commission said in its proposed rule (and that’s almost certainly not capturing Polymarket’s main, offshore exchange).

At the same time, this booming business is under fire from state gambling regulators, who claim platforms are operating in their jurisdictions illegally, and scrutiny from regulators and law enforcement over alleged insider trading. Prediction markets have joined crypto as a frequent target of criticism from Sen. Elizabeth Warren (D-MA) and other progressive critics, while also drawing opposition from some conservatives, the aforementioned state regulators, and gambling interests.

Yet this proposal comes just 40 days after the comment period ended for the CFTC’s advanced notice of proposed rulemaking for prediction markets (which received about 3,500 comments – read ours here). That’s an unusually tight turnaround by Washington standards.

We are not lawyers, and we can’t claim to have read the entire proposal (at 79,139 words, it’s almost as long as “Paradise Lost”). But our general impression is that it signals a meaningfully more open CFTC posture toward prediction markets. We wouldn’t be surprised to see the states seeking to protect their turf challenge the proposed rule in court. Then again, we remain indifferent to sports betting.

This much seems clear: under 36-year-old chairman Michael Selig, the CFTC is moving fast to set clear rules of the road for prediction markets. – Marc Hochstein

Anthropic’s Fable 5 Signals Where Frontier AI Is Heading

Anthropic on June 9 released Claude Fable 5, the first publicly available model from its new Mythos class. Fable runs on the same underlying model as the restricted Claude Mythos 5, wrapped in safeguards that make it safe for general use. Three things about how it ships break from past releases. It's free on subscriptions for only a fixed window, it requires 30-day retention on all user traffic, and it carries safeguards that either route requests to a weaker model or quietly degrade its own output.

Fable launched at No. 1 on Artificial Analysis's intelligence index, roughly five points clear of the next non-Anthropic model, and Anthropic says its edge grows on longer, harder tasks. It's priced at $10 per million input tokens and $50 per million output token, double the cost of Opus 4.8, and burns roughly twice the usage. Fable is included on Pro, Max, Team, and Enterprise plans only through June 22, after which Anthropic says use will require usage credits "due to capacity constraints," with a plan to restore subscription access once it has enough compute.

As part of the release, Anthropic now requires 30-day retention on all Mythos-class traffic across first and third-party surfaces, while saying it won't train on the data. Fable's classifiers route any request touching cybersecurity, biology, chemistry, or distillation to the weaker Opus 4.8, with the user notified. A further safeguard, disclosed in the 319-page system card, is invisible: Fable degrades its own work on frontier AI development, such as building pretraining pipelines or training infrastructure, through prompt modification and steering vectors rather than an outright block, and without telling the user.

Our Take

It can feel like the past year exhausted the AI news cycle, but the forces behind this release all point to more disruption ahead, not less.

On the market side, the supply constraint that pushed Fable behind a meter is not resolving quickly, as even Anthropic's multi-gigawatt compute deals land in 2027, so frontier capability is likely to stay expensive and rationed, with the attendant effects on who can afford it. On the capability side, days before the launch, Anthropic warned that models may be approaching the point where they can meaningfully improve themselves, noting that Claude already writes most of the code merged into its own codebase. The system card stops short of declaring that threshold crossed, but the direction is clear. A tightening compute market layered on top of models inching toward self-improvement is the backdrop every investor, builder, and policymaker should be underwriting for the coming year.

The compute squeeze is already reshaping who gets access to what, and metering Fable on consumer plans is the first sign of a broader repricing, pushed by profitability pressure ahead of fall IPOs and by compute that stayed scarce even as Anthropic's run-rate revenue climbed from roughly $9b to past $30b. Buyers are responding by spending less, and the market is splitting in two. Citadel Securities mapped that split in a recent note, "Tokenomics," pointing to Amazon pulling its token leaderboard, Microsoft cancelling Claude Code subscriptions, and a run of outsized token bills. Citadel reads the recent decline in the Silicon Data LLM Token Expenditure Index, a benchmark for the effective price of LLM usage, as users shifting to cheaper models wherever they don't need the frontier ones, and calls it the early sign of a "bifurcation in frontier vs everyday AI usage." The same divide shows up in OpenRouter's $113m raise at a $1.3b valuation, as its routing volume rose fivefold by sending each request to the cheapest capable model. A deflating commodity tier and an inflating, rationed frontier are pulling apart from each other, and that gap turns the case for open-weight models and decentralized, verifiable compute from ideological into economic.

Capability is the other force, and it cuts both ways. A model capable enough to help build its successor is also capable enough to do serious damage in the wrong hands. By Anthropic's own evaluations, Fable's underlying model is the strongest cyber model the company has built, and it sits just below its threshold for aiding novel bioweapons research. Fable itself can't be turned to those ends, since its classifiers send those requests to Opus 4.8. But the capability exists, now fenced behind a classifier that outside testers have already found ways through. Crypto has already experienced the consequences firsthand. Over the past year, a steady run of on-chain exploits has drained real money wherever attackers could find and weaponize a flaw.

The harder problem is that policy and regulation continue to lag. Alongside the Fable launch, Anthropic CEO Dario Amodei published an essay arguing that AI now needs binding rules rather than voluntary ones, including FAA-style mandatory testing that would let regulators block a model's release. The executive order the Trump administration signed on June 2 went the opposite way, building a voluntary framework that explicitly rules out any mandatory licensing or preclearance for new models. With one of the AI industry’s largest companies asking for limits the government won't impose, decisions about who gets frontier capability, and on what terms, now sit with a handful of private labs and their leadership. -Lucas Tcheyan

Other News

👀SEC proposes rescinding old rules that hold back tokenized stocks

🏦 Citi is rolling out tokenized shares of private companies

🧺 DOJ charges 2 in $389m crypto money laundering case

🔎 Trump to nominate ex-SEC chief Jay Clayton for DNI

🌯Circle debuts own wrapped BTC token, taking on wBTC and cbBTC

🔐Morpho raises $175m at $2b valuation, “largest DeFi round ever”

🚨Raydium exploited for $1.3m; treasury to compensate

🎭Fund giant Janus strikes multifaceted deal with Ethena

🤖Goldman Sachs, JPM said to mull trading compute futures

👔Banks plan tokenized deposit system (nicknames: “the bridge,” “the chain”)

Charts of the Week: Bitcoin Hasn't Bottomed Yet and the Prediction Market Duopoly Heats Up

In Galaxy Research’s new report, “Bitcoin Hasn’t Bottomed Yet. Here’s Where the Data Says It Could,” Alex Thorn uses onchain and market data to identify durable bottom indicators from prior cycles and assess their status during the current drawdown. That analysis produces a scenario ladder with a shallow ($51-54k), base ($40-46k), and harsh ($30-37k) bottoming scenarios. The chart below visualizes those levels.

Sports bets make up the majority of Kalshi's volume, so strip it out and see who actually owns the event contract business. For all of 2025, it was Polymarket and no one else. Now, that gap is gone. As of June 8, the lines have all but converged (Polymarket: ~$1.18bn, Kalshi: ~$1.0bn), meaning the regulated venue has caught up to the crypto-native incumbent on the serious markets, not just the games. For more, read our report, “The Race to Trade Everything.”