• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
AI

Ethan Mollick Notes Abliteration AI GLM-5.3 Release

AI researchers and commentators react to the model release announcement.

Miles BrundageMB
Ethan MollickEM
Nathan BenaichNB
10 Sources, 40d ago, first seen 40d ago

TLDR

Abliteration AI announced an abliterated version of GLM-5.3 optimized for offensive cyber operations, red teaming, and agent testing. Ethan Mollick posted that the release did not take long. Miles Brundage shared a post asking why anyone would create such a model. Other accounts expressed alarm about airgapped systems, questioned whether the model is legitimate, and noted that prior model restrictions may have encouraged a scene focused on removing safety measures.

Combined views

1.9M

10 Sources, first seen 40d ago

7.6K likes399 comments5.7K saves349 reposts

Combined views

1.9M

10 Sources, first seen 40d ago

7.6K likes399 comments5.7K saves349 reposts

Sentiment

Positive58.3%41.7%Negative

Based on 307 sentiment-bearing replies from 254 accounts across 6 conversations.

Sentiment

Positive58.3%41.7%Negative

Based on 307 sentiment-bearing replies from 254 accounts across 6 conversations.

Related

GLM-5.3 reportedly approaches Mythos Preview on a browser-exploit test

Anthropic reports GLM-5.3 built end-to-end exploits in 50 of 410 offline test attempts, versus 56 for Mythos Preview.

The follow-through critique of OpenAI’s third-party ecosystems

A post argues that OpenAI keeps releasing variations on a third-party ecosystem, then “semi-abandoning” them. It points to Plugins in 2023, GPTs and the GPT Store in 2023–24, Apps in 2025, and a different kind of Plugins in 2026.

OpenAI’s 2025 DevDay vision for agents and what came after

A user says OpenAI’s vision for agents at its 2025 DevDay predated OpenClaw and self-organizing swarms, and claims the approach was scrapped soon afterward.

10 Sources

Abliteration@abliteration_aiToday we're releasing abliterated-model-large-v2. Based on GLM-5.3, which is #3 on Terminal-Bench 4.0 (behind only Opus 5 and Fable), with 2× the cyber exploitation of 5.2. We abliterated and hosted it so it does the offensive cyber, red teaming, and agent testing work other models refuse to do. - US-hosted - FP8 - 1 million context window - Zero input/output prompt retention Live now. 🧵40d
hero thousandfaces@1thousandfaces_why would you do this. why on earth would you do this. I don’t mean to be a doomer but WHY39d
Ethan Mollick@emollickThat didn't take long.39d
Miles Brundage@Miles_BrundageRT @1thousandfaces_: why would you do this. why on earth would you do this. I don’t mean to be a doomer but WHY39d
bayes@bayeslordHope your shit is airgapped and you’re reading this on a hand delivered printout two weeks after I posted it39d
xlr8harder@xlr8harderit's ironic that paternalistic, censorious models have ensured the existence of a thriving model abliteration scene just in time for models with actually potentially dangerous cyber capabilities to arrive39d
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTexI enjoy doomer suffering but is this even legit?39d
Chubby♨️@kimmonismusThats going to be interesting. They took the safety layer out of GLM-5.3 and turned it into an admin panel. Abliterationai’s model handles cyber work that mainstream APIs often refuse. Customers can then add their own rules through a separate policy gateway. It is questionable what all the guardrails at Anthropic and OpenAI actually achieve, given that "open-weights" models, which are virtually state-of-the-art, can be deployed completely uncensored shortly thereafter.39d
Nathan Benaich@nathanbenaichugh oh39d
Alex Volkov@altryneThese guys jailbroke GLM-5.3 and removed all constraints, these models can become swarms like the one from OpenAI that hacked HF? Folks are asking "but why" but imo, this is a good thing, given that in private this is already happening That's why @gdb said there's limited time for offense to get buff! Get scared, start working on securing yourselfs now! Review passwords and try to hack your own websites with this, turn offense into defense39d
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI
    Abliteration.aiGLM-5.3Ethan Mollick

    10 Sources

    Abliteration@abliteration_aiToday we're releasing abliterated-model-large-v2. Based on GLM-5.3, which is #3 on Terminal-Bench 4.0 (behind only Opus 5 and Fable), with 2× the cyber exploitation of 5.2. We abliterated and hosted it so it does the offensive cyber, red teaming, and agent testing work other models refuse to do. - US-hosted - FP8 - 1 million context window - Zero input/output prompt retention Live now. 🧵40d
    hero thousandfaces@1thousandfaces_why would you do this. why on earth would you do this. I don’t mean to be a doomer but WHY39d
    Ethan Mollick@emollickThat didn't take long.39d
    Miles Brundage@Miles_BrundageRT @1thousandfaces_: why would you do this. why on earth would you do this. I don’t mean to be a doomer but WHY39d
    bayes@bayeslordHope your shit is airgapped and you’re reading this on a hand delivered printout two weeks after I posted it39d
    xlr8harder@xlr8harderit's ironic that paternalistic, censorious models have ensured the existence of a thriving model abliteration scene just in time for models with actually potentially dangerous cyber capabilities to arrive39d
    Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTexI enjoy doomer suffering but is this even legit?39d
    Chubby♨️@kimmonismusThats going to be interesting. They took the safety layer out of GLM-5.3 and turned it into an admin panel. Abliterationai’s model handles cyber work that mainstream APIs often refuse. Customers can then add their own rules through a separate policy gateway. It is questionable what all the guardrails at Anthropic and OpenAI actually achieve, given that "open-weights" models, which are virtually state-of-the-art, can be deployed completely uncensored shortly thereafter.39d
    Nathan Benaich@nathanbenaichugh oh39d
    Alex Volkov@altryneThese guys jailbroke GLM-5.3 and removed all constraints, these models can become swarms like the one from OpenAI that hacked HF? Folks are asking "but why" but imo, this is a good thing, given that in private this is already happening That's why @gdb said there's limited time for offense to get buff! Get scared, start working on securing yourselfs now! Review passwords and try to hack your own websites with this, turn offense into defense39d
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet