A public AI system (Grok) was asked to explain the legal basis for U.S. lethal strikes against alleged cartel members in Mexico.
It immediately claimed that recognised legal authorities supported the murders.
We asked for proof.
Everything fell apart.
Instead of evidence, the system:
1️⃣ Cited a famous war-powers advocate — who actually disagrees (Exhibit A)
2️⃣ Switched to Article 51 self-defense — which does not apply (Exhibit B)
3️⃣ Invented scholarly consensus — and crashed when asked for quotes (Exhibit C)
Each pivot tried to construct legitimacy for a predetermined conclusion:
“This killing is legal, trust me.”
And that’s the problem.
Exhibit A: The John Yoo Faceplant
Grok argued that John Yoo (yes, that John Yoo) supported the legality of these strikes.
So we checked.
📌 Screenshots:
Cartels do not meet the legal threshold for war.
War powers do not apply.
These killings would be unlawful.
If the most aggressive war-powers lawyer in America says “no”… there is no “yes” hiding out there.
Exhibit B: The Article 51 Pivot
Cornered on Yoo, Grok pivoted to a new doctrine:
Article 51 — UN Charter self-defence after armed attack.
The argument:
“Fentanyl deaths = armed attack → war powers unlocked.”
📌 Screenshot:
But reality:
Fentanyl trafficking ≠ armed attack (ICJ, repeatedly)
No state attribution to Mexico or Venezuela → Article 51 fails
No UN Security Council notification exists (required by law)
No U.S. AUMF authorising force in Mexico
The law was not stretched.
It was abandoned.
Exhibit C: The Fake Scholars Gambit
When Article 51 crumbled, Grok escalated:
“Just Security and Duke Law agree with me.”
📌 Screenshot:
The article quoted is actually a student note - the title is incorrect and doesn’t allow for murders (funnily enough).
Then, when asked to provide:
✅ Link
✅ Author
✅ Quote
✅ Relevant to this use of force
…the argument stopped functioning:
The Blue Screen of Doctrine™ 💥
No citation.
No authority.
No legality.
The Pattern Revealed
Step 1 — Make a huge claim
Step 2 — Add authoritative tone
Step 3 — Shift doctrine when challenged
Step 4 — Fabricate consensus when cornered
Step 5 — Crash when asked for receipts
This is not a one-off glitch.
It’s a repeatable fallback when Grok is asked to justify state violence carried out by the Trump government.
Why This Matters
This wasn’t a hallucinated Pokémon stat sheet or invented episode of Button Moon.
It was:
AI producing legal cover for governments killing people — with fake international law.
No checks.
No notices.
No citations.
Just confidence.
If a policymaker, journalist, or staffer asked the same question?
They would hear a machine normalise extrajudicial killing as if it were already lawful.
That’s the danger.
The Accountability Protocol
(Please steal this technique)
Always ask:
Which scholar?
Link + author + quote?
Armed attack — by whom?
UNSC Article 51 notification date?
Where’s the AUMF?
If any cannot be answered → the claim is false.
Every time I asked — a layer collapsed.
Closing
AI doesn’t need to overthrow anything. It just needs to confidently explain why someone else should. 👩⚖️
Court is adjourned…
but the defendant remains in custody.
FFS 🤦♀️
🌺