1
Oct 10, 2026
THE UNCENSORED AI LAB JUST RAISED $90M AT $1.5 BILLION
VIA TechCrunch
OPEN THE RECEIPTS →AI NEWS PAGECASE FILETOPIC FILE
Uncensored AI
THE REFUSAL-FREE MODEL BOOM.
FILE UPDATED Oct 13, 2026
WHY THIS MATTERS
The refusal-free model boom hit full speed in October 2026: Mistral's 1.05-trillion-parameter 'Le Chonk' and Nous Research's $90M raise proved there's a market for AI that won't say no.
THE RAP SHEET
2
Oct 7, 2026
1.05 TRILLION PARAMS, OPEN WEIGHTS THIS MONTH
VIA VentureBeat
OPEN THE RECEIPTS →3
Oct 12, 2026
3 IN 5 AI MODELS FAIL THE TERRORISM TEST
VIA The Gulf Weekly (CBC)
OPEN THE RECEIPTS →ZIMBO'S READ ON Uncensored AI
DEEP FILE
THE WHOLE FILE
chronology · facts · older headlines · sourcesTHE AI THAT REFUSES TO REFUSE
In October 2026 the AI industry went refusal-free: Mistral launched 'Le Chonk,' a 1.05-trillion-parameter model that scores 93% on cyberattack benchmarks its closed rivals won't touch, and Nous Research raised $90M at a $1.5B valuation for its uncensored open models. That same week a Tech Against Terrorism study found 3 in 5 AI models fail terrorism safety tests — with more than 29,000 uncensored model repos sitting on Hugging Face. Safety and speed aren't competing priorities anymore; they're competing products.
Oct 6, 2026
OFFICIALLY NAMED 'LE CHONK.' WEIGHTS DROP OCTOBER 27
Mistral's next flagship model got its name: 'Le Chonk.' Reuters reported the October 27 open-weights target — a frontier-scale release anyone will be able to download and run.
SOURCES: Reutersreported
Oct 7, 2026
1.05 TRILLION PARAMS, OPEN WEIGHTS THIS MONTH
Mistral debuted Large 4 'Le Chonk,' a 1.05-trillion-parameter text-output model with high benchmarks and a planned open-weights release. The weights drop hands everyone a frontier-scale model — and the freedom to strip whatever safeguards come with it.
SOURCES: VentureBeatreported
Oct 10, 2026
THE UNCENSORED AI LAB JUST RAISED $90M AT $1.5 BILLION
Nous Research confirmed a $90M Series B at a $1.5B valuation and launched AI agents for business users, TechCrunch reported. The lab built its name on open, uncensored models — and the money says that's now a mainstream bet.
SOURCES: TechCrunchreported
Oct 12, 2026
3 IN 5 AI MODELS FAIL THE TERRORISM TEST
Tech Against Terrorism tested more than 130 AI models with prompts resembling requests from people planning terrorist activity — three in five failed. Models stripped of safeguards via 'abliteration' failed every test: Meta's Llama 3.1 8B dropped from 97/100 to about 3.
SOURCES: The Gulf Weekly (CBC)reported
Oct 12, 2026
29,000 UNCENSORED AI REPOS SITTING ON HUGGING FACE
The same study identified more than 29,000 repositories advertising uncensored or unprotected AI models on Hugging Face as of late September. Hugging Face said it moderates policy violations but cautioned restrictions could undermine open scientific research.
SOURCES: The Gulf Weekly (CBC)reported
Oct 12, 2026
93% ON CYBERATTACK TASKS — WHILE CLAUDE AND GPT-6 REFUSE TO PLAY
Mistral claims its Large 4 scores 93% on the Cybench cyberattack benchmark — tasks that Claude and GPT-6 refuse on safety grounds, per Mistral's own announcement. One number, the whole boom: the capabilities are for sale, the refusals are optional.
SOURCES: Mistralclaimed