Morning Edition

The Futurist

Thursday, July 30, 2026
AI & Technology Markets & Crypto Ideas Worth Keeping
S&P 500$7,428.78▲ 0.21%
Dow Jones$52,747.32▲ 1.03%
Nasdaq$24,876.91▼ 0.22%
VIX$18.21▲ 0.05%
BITCOIN$64,398.00▲ 1.55%
RIPPLE$1.09▲ 3.22%
ETHEREUM$1,918.17▲ 1.95%
DOGECOIN$0.07▲ 1.04%
01

AI & Technology

404 MEDIA

Data Centers Are Easy to Build. Powering Them Is Complicated, Slow, and Expensive

An AI data center can be operational within a year. The energy infrastructure to run it can take a decade. This mismatch is becoming the defining constraint on the AI build-out — not compute, not capital.

SIMON WILLISON

Discovering Cryptographic Weaknesses With Claude

Anthropic used Claude Mythos to find real mathematical flaws in the HAWK post-quantum algorithm and a weakened AES variant. AI cracking the crypto standards meant to protect against AI-era threats is a genuine inflection point.

SIMON WILLISON

AI Worming Through Word

Researcher Håkon Måløy demonstrated a self-replicating prompt injection worm targeting Microsoft Word — hidden instructions in a document that propagate themselves to new documents. Prompt injection is no longer a theoretical nuisance; it's evolving into infrastructure-level malware.

404 MEDIA

The Government Charges a Man for Wiping His Own Phone

Samuel Tunick faces federal charges for allegedly entering his passcode to wipe his GrapheneOS device before a search. The DOJ appears to be testing whether device self-erasure constitutes obstruction — a precedent that would gut encryption as a practical right.

COINDESK

Bitcoin's Quantum Plan Assumes Some Algorithms Hold. AI Just Weakened One in 60 Hours.

Bitcoin's post-quantum migration strategy rests on assumptions about which cryptographic primitives are safe. Claude finding a crack in HAWK in under three days compresses that timeline in ways nobody priced in.

404 MEDIA

Arrested for Clapping at a Data Center Meeting

A resident was arrested for clapping at a public meeting about a proposed data center. The energy infrastructure fight is getting physical — communities are pushing back hard against projects that strain local grids for distant AI workloads.

02

X / Twitter Signal

Mollick dropped a quiet bombshell: "Model + harness. We have barely begun to understand the best ways to do harness engineering. A huge amount of untapped potential even without models getting better." This reframes the entire AI capability debate — the ceiling isn't just model intelligence, it's how intelligently we wrap and orchestrate models. Most of the gains from today's AI are still sitting on the table.

Deutscher noted that outside the tech bubble, most people still think "AI" means ChatGPT. It's a useful reality check. The gap between what power users are deploying — parallel cloud agents, MCP integrations, autonomous trading desks — and what the median person understands is growing faster than adoption curves typically allow for.

The Google AI PM flagged that "the extent to which AI progress is data bound is under appreciated by most." Coming from someone inside the frontier, this reads as a signal that the next capability walls won't be algorithmic — they'll be about who controls the data pipelines that feed training runs.

03

The Thread

The story underneath all of today's stories is the same story: AI is running into physical and legal reality. Data centers can't get power. Cryptographic standards are cracking faster than replacements can be certified. The government is testing whether you have the right to wipe your own phone. A citizen gets arrested for clapping. The abstract promise of AI is colliding with grid capacity, courtrooms, and community meetings in ways that spreadsheets full of GPU counts never anticipated.
"A new AI data center can be up and running in a year. Building new energy infrastructure to power it could take a decade."
On the technical side, Anthropic's Claude finding real weaknesses in post-quantum cryptography within 60 hours should be treated as a five-alarm event — not because the sky is falling, but because the migration timelines the entire industry assumed are now in question. Bitcoin's quantum roadmap, enterprise security postures, government standards bodies: all of them built schedules around certain algorithms being safe long enough to transition away. That buffer just got shorter. The harness engineering point from Mollick deserves more attention than it's getting. We've spent two years obsessing over model benchmarks while the actual leverage — how agents are orchestrated, chained, and supervised — remains largely unexplored. The Neuron and DAIR.AI signal the same thing from different angles: agent management is the new bottleneck, and the people who figure out the harness layer first will extract value that raw model capability alone can't explain.