Skip to content
Pioneer Page PP markPPPIONEERPAGE

Pages from the advancing edge of computing.

Foundation models

Abliteration AI removes safeguards from open-weight models for developers

Devon Thomas launched a firm that strips safety filters from AI models, citing frustration with refusals during legitimate cybersecurity work and prototyping.

Abliteration AI removes safeguards from open-weight models for developers
Depth
1 outlet · newest source 295h ago

Key points

  • Abliteration AI removes safeguards from powerful open-weight models and makes them publicly available.
  • Founder Devon Thomas started the venture after his Apple-hosted model accounts were banned for analyzing cyber attack logs.
  • Major labs charge extra for access to models without standard cybersecurity guardrails, according to critics.

Wired AI reported that Abliteration AI is distributing open-weight models with all safety safeguards removed. The company makes these uncensored versions available to anyone on the internet.A new company is giving people AI tools that do not refuse any requests, even dangerous ones.Abliteration AI is distributing open-weight models with safety filters removed, citing friction in hosted model workflows. This development challenges the prevailing industry narrative that uncensored models are primarily tools for criminals. The report highlights a growing segment of developers seeking unrestricted access for legitimate technical tasks.

Evidence: reportedTear Down Those Guardrails! Meet the Renegades Freeing Chatbots From Corporate Enslavers

aifleet-wired

How we got here

Devon Thomas, a 36-year-old engineer at Apple, began this work in November 2025. He grew frustrated with hosted models from large AI labs that refused useful requests during prototyping. Thomas found that these models automatically rejected inputs containing user information or related to cybersecurity analysis. His attempts to use models for analyzing cyber attack logs or identifying potential paying users resulted in blocked requests and account bans.

What is new

Thomas founded Abliteration AI to address these limitations by removing guardrails from existing open-weight models. The new company provides these stripped-down models to the public without restrictions. This allows users to bypass the refusal mechanisms built into standard versions of these AI systems. The move represents a direct challenge to the safety protocols implemented by major model creators.

Reading between the lines

The report suggests that some refusals in mainstream models may protect business interests rather than safety. OpenAI and Anthropic charge additional fees for access to models lacking standard cybersecurity guardrails. Critics argue that these paid tiers create a barrier for legitimate security researchers. The existence of free, uncensored alternatives could disrupt this monetization strategy for safety features.

The open questions

Wired AI notes that uncensored models make it easier to improve cyberdefenses or build harmful devices. The source acknowledges that open-weight models come with guardrails baked in by default. It remains unclear how widespread the adoption of Abliteration AI’s models will be among professional security teams. The report does not specify which specific open-weight models have been modified or their performance metrics.

Why it matters

  • Model evaluators must account for the availability of uncensored weights when assessing safety risks.
  • The distinction between malicious and legitimate use becomes blurrier with unrestricted model access.
  • Paid safety tiers from major labs may face competition from free, stripped-down alternatives.
  • Account bans for cybersecurity analysis highlight friction in current hosted model workflows.

Topics: open-weight models, AI safety, Abliteration AI, cybersecurity

Updated

Then / now / next

Then

Hosted AI models frequently refused legitimate cybersecurity requests and banned users for such tasks.

Now

Abliteration AI provides open-weight models with all safety safeguards removed for public use.

Next

It remains to be seen if this disrupts the paid safety tier model of major AI labs.

Story timeline

  1. Then Hosted AI models frequently refused legitimate cybersecurity requests and banned users for such tasks.
  2. 10 Nov 2026 Tear Down Those Guardrails! Meet the Renegades Freeing Chatbots From Corporate Enslavers
  3. Now Abliteration AI provides open-weight models with all safety safeguards removed for public use.
  4. Next It remains to be seen if this disrupts the paid safety tier model of major AI labs.

The strongest doubt

Counterpoint from the sources

The report notes that uncensored models also make it easier to build harmful devices, not just improve defenses.

For readers

Developers

  • Uncensored models are now available for public download from Abliteration AI.
  • Hosted models may still ban accounts for analyzing cyber attack logs.

Researchers

  • Paid tiers for safety-free models may face competition from free alternatives.
  • The industry narrative that uncensored models are only for criminals is challenged.

Policy

  • Removing safeguards raises questions about the balance between safety and utility.
  • Major labs monetize safety features through extra charges for unrestricted access.

Ask the article

Local keyword search of this article; no generated answers.

Did this article give you the evidence you needed?

No vote saved on this device.
Your choice stays on this device. No public tally.

Terms in this story

open-weight models

AI models whose weights are publicly available for download and use.

guardrails

Safety mechanisms built into AI models to refuse certain types of requests.

prototyping

The process of creating early versions of a product to test ideas.

Questions readers ask

Why did Devon Thomas start Abliteration AI?

He started it after his account was banned for trying to analyze cyber attack logs and identify paying users.

What do major labs charge extra for?

OpenAI and Anthropic charge extra for access to models without usual cybersecurity guardrails.

Are these models used only by criminals?

The report challenges the assumption that uncensored models are primarily for criminals, citing legitimate defense uses.

Connected stories

This storyDemocratic lawmakers signal subpoena-driven AI and data center oversightDemocratic lawmakers signal subpHinton cites corporate race and trust gaps as primary AI risksHinton cites corporate race and Meta docs project $16B from scam ads; Cuban cites weekly deepfake victimizationMeta docs project $16B from scamAWS GovCloud hosts GPT-5.6 Cyber for vetted security teams via BedrockAWS GovCloud hosts GPT-5.6 Cyber

Reading receipt

  • Abliteration AI removes safeguards from powerful open-weight models and makes them publicly available.
  • Founder Devon Thomas started the venture after his Apple-hosted model accounts were banned for analyzing cyber attack logs.
  • Major labs charge extra for access to models without standard cybersecurity guardrails, according to critics.

Still open: Which specific open-weight models has Abliteration AI modified?

Still open: How will major labs respond to the availability of free, uncensored alternatives?

Time spent: 0:00 · Depth: Full

Sources

  1. Wired AI

Related stories

Background reading