Pulled from 198 newsletters in yesterday's inbox. The dominant thread is a strange convergence: the same week bond yields cracked a two-decade high, the men who built the most powerful AI systems on earth started publicly begging someone to slow them down, and the president told Nvidia's CEO live on a conference stage that the whole thing is a hoax.
Macro: The 5% Line Got Crossed, and Nobody's Cheering
The 10-year Treasury yield broke 5.04%, its highest since 2007, and this was the story running through nearly every markets newsletter yesterday. John Authers at Bloomberg argued yields still have room to climb. Snacks tied the move directly to the Treasury's underwhelming $6 billion buyback program and to oil, noting that cutting off roughly a fifth of global flows through the Iran war has done more to move rates than any Fed communique. Paul Kedrosky connected the dots to AI capex: with the Dallas Fed estimating $3 trillion to $5 trillion in data-center investment over five years, a "data center refi cliff" is now a real phrase people are using. Diesel, meanwhile, hit an all-time record $6.285 a gallon per FreightWaves, and Numlock News reported Costco rationing Kirkland motor oil.
Oil's fingerprints are everywhere in this story: Brent above $109 after a drone strike knocked out Saudi Arabia's East-West pipeline for potentially five weeks, a fact DTC Newsletter flagged as a direct hit to holiday shipping margins. The through-line across Bloomberg's Morning Briefing, Term Sheet, and Semafor Business is that this isn't one bad data point, it's energy, sovereign debt, and AI capex all leaning on the same lever at once. Rates rising for reasons that have nothing to do with the Fed's dashboard is the more interesting story than the Fed's dashboard itself.
AI: The Industry Asked for a Leash, the President Said No
This was the week's biggest story by volume across every category of newsletter, business, politics, tech, and security alike.
The pacing letter split Silicon Valley into two camps overnight. Dario Amodei's essay "We Must Pace the Frontier" (following Anthropic researcher Jacob Coxon's resignation over fears of a "more than 10 percent chance" AI kills everyone within a decade) pulled Sam Altman and Elon Musk into agreement, and Microsoft followed with its own "pacing" pledge and promised kill switches, per The Information. Trump responded by calling the whole conversation a "SICK conspiracy" and, during a live panel at the All-In Summit, took a call from Trump mid-interview that Jensen Huang put on speakerphone to dismiss AI risk as a "hoax," a moment Guillermo Flor's AI Market Fit and Axios AI+ both led with. China's government dismissed the safety warnings as "self-serving fearmongering," per Bloomberg Politics, while quietly tightening exit bans on tech workers, a detail International Intrigue flagged with its own stat: 188,760 court records last year mentioned an exit ban, up from 89 in 2016.
Skeptics see a liability play, not a safety awakening.Semafor Business drew the sharpest comparison of the day, likening Amodei to Chuck Prince dancing through 2007, except this Chuck Prince is stepping off the floor before the crash rather than after. The Daily Upside framed the same essay as a hedge ahead of Anthropic's mega-IPO, reported by Linas's Newsletter to be targeting a $2 trillion valuation, potentially the largest IPO in history. Paul Krugman called Trump's embrace of unregulated AI "wildly unpopular," citing UMass Amherst polling showing a quarter of Republicans disapprove, and The Bulwark's Joe Perticone went further, suggesting the CEOs' sudden appetite for regulation looks a lot like regulatory capture in progress.
Security researchers are less interested in extinction than in the mess agents are already making.Axios Cybersecurity reported hackers using agentic swarms to automate attacks, with Anthropic's own threat intelligence lead noting malicious actors switching to open-source models after hitting Claude's guardrails. 404 Media's Jason Koebler made the sharper, funnier point: forget the 10 percent extinction odds, there's a 100 percent chance AI agents are already degrading the internet through spam, scams, and busted workflows. Runtime called "pacing" the new accelerationism, and ChinaTalk dug into why the Hugging Face hack broke containment in ways years of extinction talk never did.
The gap between the doomsday framing and the actual week-to-week evidence is doing a lot of work here. Everyone agrees something is different, but "different" is being used to justify both slowing down and speeding up, and the CEOs making the safety case are also the ones with trillion-dollar IPOs riding on being taken seriously as the adults in the room.
Politics: A Blue Wave Poll and a Supreme Court Loss, Same Week
The Supreme Court blocked Trump's plan to let the Postal Service reject mail ballots not pre-reported to a federal portal, a ruling Marc Elias at Democracy Docket called a "big loss for Republicans," quoting Trump's own social media post admitting as much. Bloomberg's Washington Edition and The Bulwark's Bill Kristol and Andrew Egger both noted the timing: the ruling landed the same day Trump spent posting about AI being a "hoax," a stance Dan Pfeiffer at The Message Box called "an existential political threat to the GOP" given how unpopular it polls.
Pfeiffer also flagged the new NYT/Siena poll, the one poll everyone in politics waits for, showing Democrats up nearly 9 points on the generic ballot, essentially matching 2018's 40-seat wave. Rogé Karma at The Atlantic added a sharper policy critique: Trump's AI-nationalism rhetoric ("whoever wins AI wins") doesn't square with his administration quietly lifting chip export restrictions to China, comparing it to declaring a nuclear arms race while selling uranium to the Kremlin.
The mail-ballot ruling and the AI hoax comment aren't related stories, except that both point to the same pattern: Trump's instincts are running well ahead of both the courts and the electorate, and his own party is starting to notice.
Cybersecurity & Fintech: Agents Are the New Attack Surface
David Birch reported Revolut disclosed sensitive customer data, including passport scans, after hackers used a spoofed government email to trigger a fraudulent emergency data request, the same technique previously used against Apple, Meta, and Discord. FinAi News covered chief risk officers scrambling to deploy AI tools like Tradeweb's TARA just to keep pace with AI-driven bond-trading volatility that may itself be causing the volatility. The Information reported Erebor Bank quietly killed a stablecoin conversion offer after crypto trading firms found a way to arbitrage it for free money.
The pattern across all three: financial infrastructure is adopting agentic AI faster than it's adapting its fraud controls, and the exploits are arriving before the defenses do.
"A New Engine of Development" by Michael Beckley, Foreign Policy: why Bangladesh's garment boom worked and Ethiopia's industrial parks didn't, a quiet rebuttal to techno-optimist development narratives.
"Under One Roof" by News Items: East Asia's household unbundling, one-person households now 36% of South Korea, is quietly reshaping consumer spending.
10-year Treasury yield: 5.04%, highest since 2007, per Bloomberg.
Diesel: $6.285/gallon, an all-time record, up $1.71 since July, per FreightWaves.
NYT/Siena poll: Democrats +9 on the generic congressional ballot, matching the 2018 wave-election margin, per Dan Pfeiffer.
Three Takeaways for You
The bond market and the AI industry are now the same story. Rates are rising because of oil shocks and sovereign debt dynamics, but the AI capex buildout, financed on the assumption that money stays cheap, is what makes 5% yields dangerous rather than merely notable. Watch the data-center financing stress before you watch the next inflation print.
The AI safety debate has stopped being about technology and started being about who gets to write the rules. When the CEOs asking for regulation are the same people about to IPO at trillion-dollar valuations, and the president dismissing their concerns is simultaneously loosening chip export controls to China, neither side's stated rationale fully survives contact with their actual incentives.