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.
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.
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.
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.
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.
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.
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.