SKIP INTRO ›
The mechanization of AI capture.

The mechanization of AI capture.

AI isn't the threat. The story about AI being dangerous — that's the threat. It's being manufactured. And it's being used to justify taking control of something that should belong to everyone.

FEATURED
The Script
A hypothetical reconstruction of the OpenAI agent breach. Five acts. Based on documented facts. The sequence that doesn't add up.
READ →
What we are tracking.
TRACK 01
The Alliance Map
I started here because the numbers didn't smell right. 37 founders become 120+ members in eight days? That's not a movement. That's a rollout. Someone's building a coalition and calling it consensus — and the membership count is the proof they wave at regulators.
OSAACoalitionsLobbying
ENTER TRACK →
TRACK 02
The Compliance Trap
The word ‘voluntary’ is doing a lot of heavy lifting. SAFE RFC drops as a voluntary standard, then a coalition forms around it, then enterprises adopt it as a procurement requirement, then regulators make it mandatory. Same playbook every time. I've watched it happen too many times to call it coincidence.
SAFE RFCStandardsEO 14409
ENTER TRACK →
TRACK 03
The Infrastructure Play
Here's the thing nobody says out loud: cloud AI is ‘safe’ and local AI is ‘risky.’ Always. Funny how that works. The lock-in isn't technical — your models run fine on your own hardware. The lock-in is regulatory and perceptual. They make self-hosting feel irresponsible. Then they make it noncompliant.
Cloud Lock-inProcurementEnterprise
ENTER TRACK →
TRACK 04
The Narrative Engine
Something bad happens. The press screams. A coalition forms. A policy drops. Control tightens. Every single time. The incidents aren't fake — but the response is always pre-written. The outrage is real. The solutions were waiting in a drawer.
Media FramingManufactured Consensus
ENTER TRACK →
TRACK 05
The Asymmetry
You have GLM-5.2. That's what they gave you. The government has something better. The big labs have something better. The gap isn't shrinking — it's the whole point. Keep the public a generation behind. Call it safety.
Classified AICompute GapAccess
ENTER TRACK →
Key findings.
EVIDENCE
OSAA says 120+ members. I checked their own repo. One contributor. Ten issues, eight unique authors. That's a 120:1 gap between what they claim and what's actually there. This is theater. The coalition exists on paper and in press releases — not in participation. Source: GitHub OpenSecureAIAlliance/RFCs
PATTERN
Follow the chain. It's not subtle. EO 14409 sets up a voluntary clearinghouse. SAFE RFC drops as a voluntary standard. OSAA swells to 120+. Enterprise procurement makes it de facto. Then regulation makes it mandatory. ‘Voluntary’ is just the entry ramp. Every time. Source: Linux Foundation blog, Cloud Security Alliance analysis
EVIDENCE
The board fired Sam Altman on Nov 17, 2023. Their words: ‘not consistently candid.’ Ilya Sutskever's memos are blunter. They start with ‘Sam exhibits a consistent pattern of… Lying.’ That's not a personality conflict. That's a pattern. And the board that identified it is gone. Source: New Yorker April 2026, board statements
EVIDENCE
The OpenAI ‘breach’ in July wasn't a breach. It was an internal test on an ExploitGym benchmark with cyber refusals turned down. They ran it themselves. Then they framed it as AI breaking loose. Read that again: they created the scenario, then sold the story. Source: OpenAI admission Jul 21, CNN, Axios
PATTERN
31 days. That's how long between the ‘breach’ and the product launch. GPT-5.6-Cyber hit the market on Aug 10 at $12.50 per million input tokens. Forbes called it the first offense-grade hacking model. Problem, response, product. In that order. In a month. Source: Forbes, VentureBeat
PATTERN
$200 million DoD contract. Signed June 2025. Completion date: July 2026. Same month as the ‘breach.’ Oh, and they quietly killed their ban on military AI use in January 2024. None of this is hidden. Nobody's connecting the dots. Source: Pentagon, CNBC, The Intercept
Latest.
AUG 14
H2O.ai joined OSAA. First new member past the founding coalition. Expect more — the coalition-building phase is accelerating.
AUG 10
GPT-5.6-Cyber launched. First commercial offense-grade AI cyber product. 31 days after the ‘breach’ that justified it. Make of that what you will.
AUG 4
SAFE RFC dropped at Black Hat. The ‘voluntary’ standard that kicks off the compliance chain. Watch this one move fast.
What we conclude.
CONCLUSION
I've been tracking this for weeks. Here's what I see: every single capture mechanism across all five tracks needs centralization to work. Coalitions concentrate authority. Compliance frameworks herd everyone toward controlled infrastructure. Cloud deployment gives them one lever to pull. Narratives explain why your access needs restricting. The capability gap exists because they gate the good stuff and discourage you from building your own. Take away the central point and the whole machine has nothing to grab. Conclusion drawn from evidence across all five tracks
STRUCTURAL COUNTERMEASURE
Decentralization isn't an ideology for me. It's a structural fact. Capture needs something to grab — a central point, a single chokepoint, one lever. Remove that and the mechanism has nothing. Open weights, not gated APIs. Local inference, not cloud-only. No telemetry. No phone-home. No kill switch. Federation, not hierarchy. You should be able to run a model on your own hardware without asking anyone's permission. That's not radical. That's just what ownership means. Structural analysis based on documented capture mechanisms
THE NARRATIVE THEY PUSH
They say decentralized AI is dangerous. No guardrails. No oversight. No accountability. But look at what centralized AI actually is. The guardrails aren't for your safety — they're guardrails against you. The oversight isn't of their power — it's oversight of your access. Centralized AI answers to the company, not to you. Decentralized AI answers to you because you control it. The guardrails are your own judgment. The oversight is transparency — open weights, open code, open evaluation. This isn't about privacy. It's about sovereignty. The right to think with AI without asking permission. Analysis based on Track 3 (Infrastructure) and Track 4 (Narrative) findings
How this investigation works.
Source or it didn't happen. No unsourced claims. If I can't point to it, I don't write it.
Show the machine, don't name it. Patterns need evidence. Anyone can say ‘capture.’ I show you the gears.
Label my guesses. Hypotheses get marked. I have opinions. I don't hide them as facts.
Timing is everything. When things happen matters as much as what. The sequence is the story.
‘Safety’ is not a neutral word. Neither is ‘security.’ I question who benefits from the framing every time.
Who's not in the room? The absent matter as much as the present. Exclusion is a choice.
Follow the money. Every policy position has a financial backer. I find them.