The 20 Percent Monitoring Bill
OpenAI says monitoring Astra costs roughly 20 percent of the compute being monitored. That figure shows effort, not effectiveness.
Author archive
Beau Henry
Under-Elicited Is a Result
Anthropic called a Claude Opus 4.6 sabotage evaluation under-elicited. Before a low score becomes safety evidence, the test needs a positive control.
The Test the Lab Never Sees
DeepMind protected a private benchmark and proprietary model weights inside one sealed evaluation. The public still needs a receipt showing what happened there.
The Price of an AI Safety Standard
AI safety assessments may become a condition of releasing frontier models. Before that happens, lawmakers should publish the method and the cost of compliance.
The Anthropic Red-Team Hour
Three outside teams tested Claude Opus 5 safeguards and reported it in hours, hours, and attempts — the same unit problem I found in the OpenAI system card in August, now in the document from the company I am applying to.
Ghostwriting for a Company That Might Be Wrong
Two years writing what a privacy company said about itself taught me that the job was never really writing, it was verification, and nobody hires for that part out loud.
Strong Opinions, Held Loosely, Written Down
Why the loosely held half of the phrase only works if you write it down first
The Configuration Is the Finding
Benchmark numbers without their configuration are anecdotes wearing decimal points. Here's what should be standard.
Which Expert Did You Beat?
How Expert Baselines Turn Methodological Choices Into Model Claims
Calibration is an Operations Problem
Where the GPT-5 System Card Separates Evidence From Judgment Well, Where It Does Not, and Why That Is an Operations Job
The Red Team Hour
The Most Honest Number in the GPT-5 System Card, and What It Leaves Out
A System Card is a Ledger
A close reading of the GPT-5 System Card asks where evidence ends, judgment begins, and safety claims lose precision as they move through the document.
Stop Chasing the Room
Earn Authenticity; Don't Chase a Seat at the Table
Beau's Reading List
Books for Success
The Calibration Standard
How to Calibrate Your AI Helper
Mastering Travel Privacy: Essential Strategies for the Modern Traveler
Considerations for Maintaining Your Digital Privacy While Traveling
Strong-Ass Passwords
Define Strong-Ass Passwords and How to Create Them
How Do I Know You Are Who You Say You Are? Or, How MFA Works
Turn on 2FA for All of Your Accounts, Please
Getting Your Shit Together (Online Privacy for People Who Give a Fuck)
Small Steps You Can Take to Gain Control of Your Privacy
The Importance of Testing Your Code
blog post
AB Testing Blog Post
product market fit
GL-iNet Beryl (GL-MT1300) Travel Router Review
gl-inet travel router beryl gl-mt1300 overview privacy setup vpn wireguard open source
Redis vs Dynamo DB
How to Pick a Fight