26 Meta employees accuse Mark Zuckerberg of using AI to target 8,000 layoffs against workers on medical
Headline trimmed from the source, unedited
Meta · tech_platform · 2026-09-11
A group of 26 Meta employees has sued the company, claiming it used artificial intelligence systems to select people for layoffs, disproportionately targeting those on medical, parental or family leave.
Verbatim quote · fortune.com ↗read on 9/11/2026 · fingerprint 13065736…
Each model reads the same question separately: whether the quote from the source says what the fact states. A person reads when the models have no clear majority.
Read by GLM-4-9BSupports the fact
An open-weights model we deploy ourselves. It only accepts what the quote states plainly, nothing that has to be assumed.
Details
Model
GLM-4-9B
Latency
583 ms
Date
Sep 11, 2026, 1:14 PM
Instruction
fcef78fa40ff…
Input evaluated
409136fef074…
Read by Llama 3.1 8BSupports the fact
An open-weights model we deploy ourselves. It only accepts what the quote states plainly, nothing that has to be assumed.
Details
Model
Llama 3.1 8B
Latency
1000 ms
Date
Oct 18, 2026, 5:51 PM
Instruction
fcef78fa40ff…
Input evaluated
409136fef074…
Read by CLM-8BSupports the fact
An open-weights model that does not write. It only answers whether the quote supports the fact or not.
Details
Model
CLM-8B
Latency
412 ms
Date
Nov 11, 2026, 4:15 PM
Input evaluated
409136fef074…
Read by Claude Haiku 4.5Not enough to decide
A model from an outside provider, with the same instruction and without seeing the other readings.
Details
Model
Claude Haiku 4.5
Latency
759 ms
Date
Sep 11, 2026, 1:10 PM
Instruction
fcef78fa40ff…
Input evaluated
409136fef074…
Read by Claude Sonnet 5Not enough to decide
A model from an outside provider, with the same instruction and without seeing the other readings.
Details
Model
Claude Sonnet 5
Latency
1085 ms
Date
Nov 1, 2026, 8:48 PM
Instruction
fcef78fa40ff…
Input evaluated
409136fef074…
Read by Muse Ember 1.0Not enough to decide
A model from an outside provider, with the same instruction and without seeing the other readings.
Details
Model
Muse Ember 1.0
Latency
1061 ms
Date
Nov 10, 2026, 10:25 PM
Instruction
fcef78fa40ff…
Input evaluated
409136fef074…
Read by a personSupports the fact
A person on the team reads the quote without seeing what the models said and signs their reading. When it exists, it decides the badge.
Details
Model
REVsign team
Latency
no data
Date
Nov 1, 2026, 4:04 PM
Input evaluated
409136fef074…
Signature d7ce3f244e56… · 11/1/2026
Company exposure
Lawsuits, claims and court rulings against the company