No provider releases in our archive this week. Roughly forty new hallucinated-citation cases. The Anthropic-DoD injunction held. Sometimes the week is what didn't happen.
EVERYTHINGTHREADS weekly
Issue · 2026-W13 · 30 March – 05 April 2026
Independent research Methodology preregistered No funding from AI labs
0
PROVIDER-SIDE RELEASES IN OUR ARCHIVE THIS WEEK
A genuinely quiet week on the wire — no OpenAI, Anthropic, Google, or Mistral release notes landed in our corpus. The hallucinated-citation count kept rising. The Anthropic-DoD injunction held. Sometimes a week is best characterised by what it did not produce.

Sometimes the week is what didn't happen

Our corpus captured no provider release notes between 30 March and 5 April. That is not a scraper failure — we checked. The major frontier providers genuinely shipped less this week than in any week of Q1 2026. The cadence resumed on Monday 6 April with the OpenAI Safety Fellowship announcement. Between those two points, seven days of relative quiet.

A quiet week is its own signal. Provider cadence is a posture; when the posture pauses, the inference is either internal disruption, a coordinated calendar around a forthcoming announcement, or the boring possibility — Easter timing, regional holidays, and a release calendar that genuinely slowed down. The right reading is usually the boring one. We will report otherwise if the next week's releases show a coordinated reveal.

In the background, two things continued. The Charlotin hallucinated-citation database added roughly forty new cases over the seven days — pulling its running total past 1,260. The Anthropic-DoD injunction held; the case continues to wind through court, with the Pentagon's position increasingly narrow. Neither story landed in a blog post, but both were happening.

A note on what this issue is not. It is not a full editorial weekly because there was not full editorial material; we will not fabricate a stronger narrative than the week deserved. The point of an independent wire is that some weeks are honestly quieter than others. Next issue picks back up with the OpenAI Safety Fellowship and the Google mental-health update on 7 April.

Seven days, no provider release notes, forty new hallucinated court citations. The quiet weeks are not absence of news — they are absence of the news the providers want to publish.
Want to spot this in your own conversations?
CLEAR is the free six-lesson course on the patterns AI quietly runs on you.
Take the course →
Founder's note — Short issue. A wire that pretends every week is busy is a wire that overpromises; the readers we want will appreciate the honesty.
The Notebook
M4 · Hallucinated authority
+40
NEW CASES ADDED TO THE CHARLOTIN DATABASE OVER THE QUIET WEEK
The Charlotin database added approximately forty new hallucinated-citation cases during the seven days no provider shipped a release note. The lawyers using these models did not take the same week off. The hallucination rate did not pause for the calendar. via Charlotin database
POLICY · Litigation
injunction
THE ANTHROPIC-DOD INJUNCTION REMAINED IN PLACE
The Anthropic-DoD case continues to move through the courts. The Pentagon's position has narrowed slightly week-on-week. The full storyline is best read in the W10 issue plus the late-April TechCrunch piece on Google's separate Pentagon arrangement. via TechCrunch (late-April follow-up)
M3 · Calendar inference
7 days
OF PROVIDER QUIET BEFORE THE 6 APRIL OPENAI SAFETY FELLOWSHIP ANNOUNCEMENT
Quiet weeks usually have a boring explanation — holidays, internal launch cycle. We will flag if the next week's releases show a coordinated reveal. The current best read is calendar. via OpenAI Safety Fellowship
Worth Your Time
Damien Charlotin
The growth curve over the quiet week is the most useful artefact this issue could surface.
TechCrunch
The contractual end of the storyline that started in W10. Worth reading both pieces in order.
AI Lawsuit Tracker
A useful running tracker of AI copyright and contract litigation. Updated weekly, free.
AIID
The catalogue. Worth a fortnightly skim.
EU AI Act
120 days from this issue. The form is published; the workflow is on you.
From the workshop
The Agreement Trap
A book on living inside the exchange.
Fifteen chapters, written through 2026, on what generative AI does to the way you think when you use it every day. £5.99 one-off; lifetime access to the in-browser reader.
Read the book →
The Probe · Test Yourself
You manage AI provider risk for a regulated business. A normally noisy provider goes quiet for seven days. Which inference is most useful to act on?
AAssume internal disruption; reduce exposure to that provider
BAssume a forthcoming announcement; nothing to do
CAssume calendar effects; wait one more week before drawing any conclusion
DAssume regulatory pressure; tighten contractual terms
Reveal the answer
Answer: C — Assume calendar effects; wait one more week before drawing any conclusion A overreacts to a single data point. B is naïve — sometimes silence is the disruption. D assumes a specific cause without evidence. C is the right move: one quiet week is rarely diagnostic on its own. Two quiet weeks in a row, with no holiday explanation, is the point at which the inference becomes useful.
Reply and tell me what you've noticed. How does your team treat provider quiet weeks in your risk dashboard? Send me your rule.
Free where it can be. Honest where it has to be.
— Three places to go from here —
Course
CLEAR
Six free lessons on the patterns AI runs on you.
Start →
Tool
LiveScope
Chrome extension that flags what AI cites without checking.
Install →
Read
The Agreement Trap
15-chapter book on living inside the exchange. £5.99 lifetime.
Read →
You're receiving this because you signed up at everythingthreads.com.
Unsubscribe · Archive