Category: Business

CoW Swap, an Ethereum-based decentralized exchange aggregator, warned users on Tuesday to avoid using the protocol, disclosing that its front-end interface had been compromised.

“We are now actively working to resolve the situation,” the project frequently used by Ethereum co-founder Vitalik Buterin said in a post to X. “The CoW Protocol backend and APIs were not impacted, but we have paused them temporarily as a precaution.”

CoW Swap indicated that attackers had gained control of the website domain that users typically visit before engaging with the protocol. That gave bad actors the opportunity to direct users to a different website where funds could be stolen through the approval of malicious transfers.

Although the compromise didn’t affect CoW Swap’s underlying smart contracts, the protocol appeared to remain frozen three hours after the attack was divulged. Meanwhile, users on Discord reported losses within the project’s official server.

“I don’t know what to do anymore,” said one user who claimed that they lost more than $50,000 via CoW Swap’s compromised front end. “I have no money at all.”

Despite apparent frustrations, the scope of losses sustained wasn’t immediately clear. 

A pseudonymous member of the CoW Swap team who goes by MooKeeper told Decrypt that reports are actively being investigated and verified. They added that a more complete assessment would be released tomorrow or later this week.

“We have evidence that a small number of users signed malicious approvals for very small amounts,” MooKeeper added.

Still, a noted cybersecurity researcher who goes by Vladimir S. on X said that around $500,000 worth of digital assets had been “drained from a few addresses so far.”

Martin Köppelmann, co-founder and CEO of decentralized infrastructure provider Gnosis, noted in a post to X that the attack’s scope appears limited. He said that users are potentially affected only if they approved interactions with CoW Swap within the past few hours.

Websites that try to trick users by mimicking established DeFi projects aren’t entirely uncommon. Last year, for example, Curve Finance suffered its second DNS hijack. The first one, which took place in 2022, resulted in $570,000 in losses for users.

Buterin, who has swapped notable amounts of Ethereum for stablecoins using CoW Swap this year, had engaged with the protocol as recently as a week ago, data from on-chain analytics firm Arkham Intelligence showed. In 2024, he also used the decentralized exchange aggregator to offload holdings of a meme coin modeled on a baby pygmy hippo from Thailand.

President Donald Trump is set to host another in-person event for his meme coin holders later this month at Mar-a-Lago, but in perhaps a sign of times, the price to attend is considerably lower than it was last year.

The team behind President Trump’s Solana-based token, which trades as TRUMP, finalized the list of attendees on Tuesday, with the guest list based on a leaderboard tied to token holdings and other factors.

Last time around, a select group of attendees attained “VIP” status by each amassing millions of dollars’ worth of the tokens. But earning the VIP label for President Trump’s latest event, a luncheon and conference to be held for top holders, cost roughly 90% less this time around—down from approximately $3 million to around $300,000.

VIP status, the top level of access for the event, grants attendees a private “meet and learn” experience with “superstar” attendees like Tether’s Paolo Ardoino, retired boxer Mike Tyson, and Ark Invest’s Cathie Wood. That bonus reward, beyond access to the luncheon with President Trump, was granted to the top 29 registered TRUMP holders who earned points based on time-weighted holdings of the token from March 12 to April 14.

Based on available on-chain holder data compared to the Trump meme team’s final leaderboard, at least one account that earned VIP status and ranked 12th overall—a Solana address beginning in “Ciz7y”—never held more than 114,212 TRUMP tokens during the more than month-long snapshot period.

That amount, now worth around $318,000, peaked at about $473,000 when the token traded at $4.14 on March 15. 

Other VIPs, like Solana address “2ciorM,” transferred in around 102,000 TRUMP tokens on March 12 and made no other moves. Those tokens, now valued at $283,000, were good enough to rank the address 17th, well above the VIP cutoff. 

Those sums are dramatically smaller than those seen for last year’s initial exclusive event for Trump meme coin holders, where the top finalists each held millions of dollars’ worth of tokens to earn VIP status, which conferred exclusive benefits. The final VIP holder for last May’s event, ranked 25th, had nearly $3 million worth of TRUMP tokens around the time of the snapshot. 

At that time, gaining VIP access to the first TRUMP meme coin gala required around $4.78 million worth of the president’s official meme coin, on average. The top holder—an account labeled “Sun” that Tron founder Justin Sun confirmed was his—buoyed the group by possessing around $19 million in TRUMP at the time of the snapshot last year.

The Tron founder, who earlier this week publicly bashed the Trump family’s World Liberty Financial business—a separate crypto venture, was again the top holder on this year’s meme coin leaderboard.

This time, though, the value of his account is only around $9.3 million, as the TRUMP token’s price has fallen precipitously from its highs. He maintains a balance of around 3.3 million TRUMP tokens, or 1.4 million more tokens than the first time around.

As it stands, the official TRUMP token was recently trading around $2.80, down 29% in the last month and now 96% off its all-time high of $73.43 achieved shortly after launch. The token briefly jumped around 40% after the announcement of the latest in-person event, but has since erased those gains. 

By comparison, during last year’s exclusive event, the token closed the snapshot day trading around $13.08—about 370% higher than its current trading price. 

This time around, token holders were also awarded 10 points for every $1 spent on purchases of Trump-branded merchandise—like hats and watches—during the token-holding period, which is something that can’t be accounted for with on-chain data.

That bonus could have moved the needle for some prospective VIPs, but otherwise, it appears that holders needed to spend far less on TRUMP tokens this year to gain special access.

Last year’s TRUMP meme coin dinner drew scrutiny from Democratic lawmakers, who demanded an ethics probe into the invitation. Famously, Senator Elizabeth Warren called the event an “orgy of corruption,” alleging that people could use it to buy access to the president.

This year’s Mar-a-Lago event will require attendees to pass a background check, and ensure they are not residents of a KYC-watchlist country. According to the terms and conditions, the event, slated for April 25, may be cancelled for any reason, and does not guarantee the attendance of President Trump. 

If Trump can not attend, or the event does not take place, then attendees will receive a Trump-themed NFT. 

President Donald Trump’s nominee to lead the Federal Reserve disclosed a vast fortune Tuesday worth well over $100 million, which includes numerous investments in the crypto sector and other emerging tech startups.

Kevin Warsh, a former Fed governor with deep Wall Street ties, was required to disclose his personal finances as part of his Senate confirmation process. His filing reveals a significant net worth. Warsh has $100 million parked in a single investment fund, for instance—one of dozens of investments and income streams the former banker was required to make public.

In his previous stint at the Fed, Warsh played a key role in the historic bank bailouts that followed the 2008 financial crisis. His current investments run the gamut from traditional finance to emerging technology firms, including several in the crypto sector.

Among Warsh’s listed investments, for instance: blockchain network Solana, yield-focused Ethereum layer-2 network Blast and Optimism, Ethereum DeFi lending protocol dYdX, NFT company Dapper Labs, and Polychain, a crypto venture firm. His other crypto investments include Bitcoin trading platform Flashnet, Ethereum developer platform Tenderly, and DeSo, an on-chain social media startup.

The Fed Chair nominee has also invested in a slew of emerging tech ventures, including Contraline (a “reversible male contraceptive solution”), Cionic (“bionic movement-enhancing wearable clothing”), and Arc Boats, an electric boating company.

Warsh further disclosed a slew of investments in AI-focused companies—plus exposure to crypto-fueled prediction market juggernaut Polymarket, and Elon Musk’s SpaceX, which is gearing up to launch a potential record-breaking IPO.

Though Warsh appears to enjoy support on Capitol Hill, his path back to the Fed is far from simple. The Trump Justice Department is currently pursuing a criminal investigation of sitting Fed Chair Jerome Powell—a longtime thorn in the side of the president—and key senators have signaled they will refuse to advance Warsh’s nomination until that investigation is resolved.

Though the Justice Department’s case against Powell has incurred major setbacks, the Trump administration appears intent on continuing to pursue the matter.

If you’ve been following the local AI scene, you probably know Qwopus—the open-source model that tried to distill Claude Opus 4.6’s reasoning into Alibaba’s Qwen, so you could run something resembling Opus on your own hardware for free. It worked surprisingly well. The obvious catch: Qwen is a Chinese model, and not everyone is comfortable with that.

Jackrong, the same pseudonymous developer behind that project, heard the feedback. His answer is Gemopus—a new family of Claude Opus-style fine-tunes built entirely on Google’s open-source Gemma 4. All-American DNA, same idea: frontier-level reasoning, running locally on hardware you already own.

The family comes in two flavors. Gemopus-4-26B-A4B is the heavier option—a Mixture of Experts model that has 26 billion total parameters but only activates around 4 billion during inference, which means it punches well above its weight on constrained hardware.

Parameters are what determine an AI’s capacity to learn, reason, and store information. Having 26 billion total parameters gives the model a huge breadth of knowledge. But by only “waking up” the 4 billion parameters relevant to your specific prompt, it delivers the high-quality results of a massive AI while remaining lightweight enough to run smoothly on everyday hardware.

The other is Gemopus-4-E4B, a 4-billion parameter edge model engineered to run comfortably on a modern iPhone or a thin-and-light MacBook—no GPU required.

The base model choice matters here. Google’s Gemma 4, released on April 2, is built directly from the same research and technology as Gemini 3—the company said so explicitly at launch. That means Gemopus carries something no Qwen-based fine-tune can claim: The DNA of Google’s own state-of-the-art closed model under the hood, wrapped in Anthropic’s thinking style on top. The best of both worlds, more or less.

What makes Gemopus different from the wave of other Gemma fine-tunes flooding Hugging Face right now is the philosophy behind it. Jackrong deliberately chose not to force Claude’s chain-of-thought reasoning traces into Gemma’s weights—a shortcut most competing releases take.

His argument, backed by recent research, is that stuffing a student model with a teacher’s surface-level reasoning text doesn’t actually transfer real reasoning ability. It teaches imitation, not logic. “There is no need for excessive imagination or superstitious replication of the Claude-style chain of thought,” the model card reads. Instead, he focused on answer quality, structural clarity, and conversational naturalness—fixing Gemma’s stiff Wikipedia tone and its tendency to lecture you about things you didn’t ask.

AI infrastructure engineer Kyle Hessling ran independent benchmarks and published the results directly on the model card. His verdict on the 26B variant was pretty favorable. “Happy to have benched this one pretty hard and it is an excellent finetune of an already exceptional model,” he wrote on X. “It rocks at one-shot requests over long contexts, and runs incredibly fast thanks to the MOE (mixture of experts) architecture.”

The smaller E4B variant passed all 14 core competence tests—instruction following, coding, math, multi-step reasoning, translation, safety, caching—and cleared all 12 long-context tests at 30K and 60K tokens. On needle-in-haystack retrieval, it passed 13 out of 13 probes including a stretch test at one million tokens with YaRN 8× RoPE scaling.

The 26B extends natively to 131K context and all the way out to 524K with YaRN, which Hessling also stress-tested: “It also crushed my simple needle-in-the-haystack tests all the way out to an extended context of 524k!”

On edge hardware, the E4B is genuinely fast. Jackrong reports 45–60 tokens per second on iPhone 17 Pro Max, and 90–120 tokens per second on MacBook Air M3/M4 via MLX. The 26B MoE architecture means it offloads gracefully on unified memory systems or GPUs with under 10GB of VRAM. Hessling called it his daily driver recommendation for VRAM-starved setups.

Both models are available in GGUF format, which means you can drop them straight into LM Studio or llama.cpp without configuration. The full training code and a step-by-step fine-tuning guide are on Jackrong’s GitHub—same pipeline he used for Qwopus, same Unsloth and LoRA setup, reproducible on Colab.

Gemopus is not without its rough edges. Tool calling remains broken across the entire Gemma 4 series in llama.cpp and LM Studio—call failures, format mismatches, loops—so if your workflow depends on agents using external tools, this is not your model yet. Jackrong himself calls it “an engineering exploration reference rather than a fully production-ready solution,” and recommends his own Qwopus 3.5 series for anyone who needs something more stable for real workloads.

And because Jackrong deliberately avoided aggressive Claude-style chain-of-thought distillation, don’t expect it to feel as deeply Opus-brained as Qwopus—that was a conscious tradeoff for stability, not an oversight.

For those who want to go deeper into Gemma fine-tuning for reasoning specifically, there is also a separate community project worth watching: Ornstein by pseudonmyous developer DJLougen, which takes the same 26B Gemma 4 base and focuses specifically on improving its reasoning chains without relying on the logic or style of any specific third party model.

One honest caveat: Gemma’s training dynamics are messier than Qwen’s for fine-tuners—wider loss fluctuations, more hyperparameter sensitivity. Jackrong says so himself. If you need a more battle-tested local model for production workflows, his Qwopus 3.5 series remains more robustly validated. But if you want an American model with Opus-style polish, Gemopus is currently your best available option. A denser 31B Gemopus variant is also in the pipeline, with Hessling teasing it as “a banger for sure.”

If you want to try running local models on your own hardware, check our guide on how to get started with local AI.