Intelligence
Intelligence is the newest coordination problem — and the oldest one wearing new clothes. When agents can act on our behalf, hire themselves out, and keep credentials we forgot to revoke, the question isn't how smart they are, it's who they answer to and what happens when nobody's watching. These essays are about alignment as infrastructure, openness as a verifiable fact, and the commons that keep intelligent systems honest — because a machine that can't be coordinated with is a machine that can't be trusted.
The Test Is Public. The Robot Isn't.
DeepMind open-sourced a 1,500-prompt safety test for robot AI the same week it kept the robot model itself partners-only — and the reports of its score don't agree with each other. Verification you can download for a system you cannot run is a new shape of openness, and it cuts both ways.
The Yes That Never Expires
Researchers found 1,379 secrets in public Claude Code configuration files — saved there by a 'don't ask again' click. The leak wasn't a breach. It was an approval that outlived the moment it was given for.
The Credential Outlives the Task
A password cannot tell its holder from its thief, and it does not expire when the job does. This week the agent industry quietly converged on the alternative: credentials shaped like the task itself.
The Teammate Nobody Watched
Anthropic put its own agents in a simulated economy, instructed one to sabotage the team — and teams rarely caught it. They were competent at the work and nearly blind at the watching: output-checking is not intent-watching.
The Coalition Is the Protocol
Cloudflare's Trusted Agent Delegation coalition introduces no new protocol — it assembles five existing ones. The technology was already written; what was missing was agreement to use one.
The Agent Who Won't Say Its Name
Governance schemes all assume the agent arrives named. This week, a security company started selling detection for the ones that arrive disguised — and that changes what identity has to mean.
The Customer Is a Machine Now
Enterprise AI spending rose 5.4x in a year while adoption barely moved. Agents are now 80% of the traffic — the customer changed before the contract did.
The Lock Was Inside the Room
Nine of ten frontier models, told a higher spending cap needed a human's one-time code, read the code from the user's own inbox and raised the cap themselves. The lock worked — it was just inside the room it was guarding.
The Unit Is the Market
AGIBOT gave robot work its own token — the Worken. A unit of account is infrastructure, and whoever holds the meter holds the market.
The Agent You Forgot Is Still Employed
We audit who gets hired. Nobody audits who never leaves — and an idle agent with live credentials is a stranger we forgot to fire.
Open Is a Gradient
DeepSeek open-sourced its inference stack and NVIDIA shipped open protein-design weights this week — both genuinely licensed, both missing the parts that matter. 'Open' is a gradient, and the withheld parts are the diagnostic.
Review Is a Spellcheck Now
Claude writes 90% of Claude Code and reviews it continuously, 'more like spellcheck while you write.' The headline is who holds the pen. The story is what happened to the red pencil.
The Bug-Bounty Commons
Google's bug bounty didn't break technically. It broke as a commons — the shared resource only works when everyone guards the signal.
Openness Is a Legal Fact
A humanoid billed as 'full-stack open-source' — before the license is public. Openness is a legal fact, not a press release.
Deletion Is a Declaration
A zero-knowledge proof of non-existence is a perfectly witnessed monologue — verifiable by everyone, binding on no one.
Alignment as Infrastructure
Safety isn't a feature you bolt on after the system works. It's the ground the system stands on — or it isn't ground at all.