The AI Czar Who Once Sued Ripple
President Trump announced a "Super Intelligence Force" to coordinate federal AI policy, and Director of National Intelligence Jay Clayton is first among its listed leaders — widely reported as the administration's new AI czar (Decrypt).
The résumé is the story. As SEC chair in Trump's first term, Clayton brought 57 crypto cases and sued Ripple over XRP in one of his final acts. He later advised crypto firms, then oversaw the prosecution of Tornado Cash developer Roman Storm as interim U.S. attorney in Manhattan.
Joining him: FTC Chairman Andrew Ferguson, Under Secretary of War for Research and Engineering Emil Michael, and OPM Director Scott Kupor. The force reports to Trump and chief of staff Susie Wiles, and follows what Trump called the "Historic White House Accord on Super Intelligence."
The man who built crypto's regulation-by-enforcement playbook now gets to write the AI rulebook. If you believe institutional memory shapes policy, expect the same instinct: regulate first, define the terms later.
Near Intents Gets Its $3.8M Back After a 48-Hour Staredown
Cross-chain swap service Near Intents recovered the full $3.8 million drained in Thursday's exploit, general manager Alex Shevchenko said Friday, and the team has ended its investigation (Decrypt).
- The return came a day after Shevchenko publicly posted return addresses and told the attacker, "We have identified you, sir," with a 48-hour window framed as a last chance at responsible disclosure.
- An on-chain message appearing to come from the exploiter said "we were in the wrong" and urged others to use bug bounties.
- The bug lived in how Near Intents' Omni deposit/withdrawal layer interacted with its main smart contract. Service was halted Thursday; the team pledged full user compensation and reported the incident to law enforcement.
- Blockchain sleuth ZachXBT said the stolen funds were sent to KuCoin and bridged to Bitcoin.
"We have identified you, sir" is a great line, but the recovery is the anomaly, not the template. Most exploiters don't send a contrite on-chain note — they send a mixer.
Grok Is Running 60% of Coinbase's AI Trading
According to coverage from Yahoo Finance and 24/7 Wall St., Grok now holds roughly a 60% share of AI-driven trading activity on Coinbase. The reports don't detail which coins are being bought with any rigor.
Meanwhile, Cathie Wood told investors to follow AI agents that spend money, while a panel warned about control — as aggregated by NewsCord.
A 60% share for one model inside one exchange's AI trading is not a market signal — it's a default setting. Concentration in a black box is a risk, not a feature.
Open Models Are Eating Enterprise Tokens
Per the Financial Times via Techmeme, mentions of open models in US earnings calls surged 6x year-over-year, with open models hitting 56% of Vercel's tokens in August and 40% of AT&T's AI workloads.
That's the demand side of decentralized compute in plain English: enterprises are routing real volume to open weights, which is exactly the workload set that distributed inference and GPU markets want to serve.
Related infrastructure money kept moving:
- PicoJool raised a $27.5M Series A led by Socratic Partners for AI data center interconnects built on vertical cavity surface emitting lasers (Techmeme).
- Blackstone, OpenAI, QTS, and SoftBank partnered with key US unions on the American Infrastructure Alliance, aiming to set data center standards in 2027 (Techmeme).
- Sources told CNBC that OpenAI offered ~$100M to invest in Hugging Face before Nvidia's $13B acquisition, with talks falling apart; AMD and Salesforce also held talks (Techmeme).
Note the pattern: the money is converging on compute, interconnects, and data centers — the physical layer. Tokens are cheap; watts and lasers are not.
Research Desk: Agents That Can't Be Trusted Yet
Two papers worth skimming if you're building agent-adjacent anything:
- DABstep, a data-agent benchmark built by Adyen and Hugging Face, found the most capable reasoning agents hit only 16% accuracy on 450+ real-world data analysis tasks (Hugging Face).
- A new prompt-injection detection framework for LLM email assistants models attacks as a chain rather than binary text classification, scoring a mean F1 of 0.406 under a strict threshold versus 0.216 for the strongest pretrained detector tested (arXiv).
The paper's blunt finding: random train/test splits substantially overestimate robustness under distribution shift. Agents look better in the lab than they behave in the wild.
Elsewhere in Crypto
- The Independent Community Bankers of America sued the OCC in D.C. federal court, challenging its authority to grant national trust bank charters to crypto firms and seeking to void a March 2026 rule and Interpretive Letter No. 1176. CEO Rebeca Romero Rainey called the charters a "side door into the banking system" (Decrypt).
- Japan added Russian exchange Garantex to its Russia sanctions list, alongside 33 organizations, nine individuals, and 35 "shadow fleet" vessels (CoinTelegraph). TRM Labs has previously argued such sanctions may be ineffective because targeted entities prepare contingency plans in advance.
- Strategy bought another 1,665 BTC for ~$142.7M at an average of $85,681, bringing holdings to 847,666 BTC — over 4% of the 21 million supply cap (The Block).
- Scammers drained $2M in ETH by impersonating the GIWA network and fooling DYORSWAP (CoinTelegraph).
Closing Take
The AI-crypto overlap this week was mostly about who holds the levers: a crypto-enforcement veteran now running AI policy, one model dominating one exchange's AI trading, and enterprises quietly routing real workloads to open weights. The decentralized-compute thesis isn't proven by any of it — but the demand curve underneath it is finally legible. Watch the policy seat, not the price charts.
Not financial advice. Verify everything; trust nothing you read on-chain or in a press release.
Social Pulse
Social chatter is thin and mostly on-chain trivia: one tracker flagged two long-dormant Bitcoin whales testing a tiny transfer after 13 years. No meaningful AI-crypto sentiment surfaced in the curated posts.
- @lookonchain — Two Bitcoin OGs who bought 1,346 BTC 13 years ago at
$178 average moved just 0.0005 BTC ($43) as a test transaction, sitting on a claimed 479x return. (source)