How much funding has Abliteration.ai raised?

Abliteration.ai has raised 0 in total funding. Their latest funding round was Bootstrapped.

Where is Abliteration.ai headquartered?

Abliteration.ai is headquartered in Palo Alto, United States. The company has a team of 1-10 people.

What does Abliteration.ai do?

Abliteration.ai operates in the AI in Cybersecurity sector. Startup that commercializes the removal of AI safety guardrails, hosting and selling API/browser access to abliterated (guardrail-stripped) open-weight models such as Z.ai's GLM-5.3 for red-teaming and offensive cybersecurity work. Hosted, OpenAI/Anthropic-compatible access to open-weight models with refusal behavior removed, so security teams can reproduce adversarial model behavior for red-teaming, offensive cyber, trust & safety, synthetic data generation, and ML research without downloading weights or securing their own compute.

When was Abliteration.ai founded?

Abliteration.ai was founded in 2025. It is based in Palo Alto, United States.

How much funding has Abliteration.ai raised?

Abliteration.ai has raised 0 in total funding.

Who is the CEO of Abliteration.ai?

Devon (co-founder; last name withheld at his request as he remains employed at another firm) is the CEO of Abliteration.ai. The company was founded in 2025.

Where is Abliteration.ai located?

Abliteration.ai is headquartered in Palo Alto, United States.

What funding stage is Abliteration.ai at?

Abliteration.ai is currently at the Active stage.

How many employees does Abliteration.ai have?

Abliteration.ai has approximately 1-10 employees.

Abliteration.ai

Category: AI in Cybersecurity

Startup that commercializes the removal of AI safety guardrails, hosting and selling API/browser access to abliterated (guardrail-stripped) open-weight models such as Z.ai's GLM-5.3 for red-teaming and offensive cybersecurity work. Abliteration.ai was founded in 2025. The company is led by Devon (co-founder; last name withheld at his request as he remains employed at another firm). Based in Palo Alto, United States. Team size: 1-10. Total funding raised: 0. Latest round: Bootstrapped.

Founded: 2025
Headquarters: Palo Alto, United States
Team size: 1-10
Total funding: 0

Value proposition

Hosted, OpenAI/Anthropic-compatible access to open-weight models with refusal behavior removed, so security teams can reproduce adversarial model behavior for red-teaming, offensive cyber, trust & safety, synthetic data generation, and ML research without downloading weights or securing their own compute.

Products and solutions

Hosted abliterated open-weight models (e.g. GLM-5.3 as abliterated-model-large-v2 / Abliterated Large v2; earlier GLM-5.2 as abliterated-model-large) via web browser and API, OpenAI Chat Completions and Anthropic Messages API compatibility, Policy Gateway (policies-as-code with allow/refuse/rewrite/redact/escalate outcomes streamed to Splunk/Datadog/Elastic/S3), Training-data generation tool, Vercel AI SDK / LangChain providers

Unique value

Turns the underground "abliteration" technique (removing a model's refusal vector) into a frictionless commercial service, letting defenders access guardrail-free frontier open-weight models without local compute, while offering a policy-as-code moderation layer for governance.

Target customer

Enterprise security teams, AI red-teamers, trust & safety researchers, ML researchers, and defense/government contractors; early-stage red-teaming startups in the UK and Europe that serve banks, airlines, and critical-infrastructure operators.

Industries served

Cybersecurity (offensive security, red teaming, agent testing), trust & safety, ML research, defense/government, synthetic data generation.

Technology advantage

Uses refusal-vector ablation to strip guardrails from open-weight models while aiming to preserve reasoning, coding, and agentic capability; drop-in OpenAI/Anthropic API compatibility; built-in Policy Gateway for enterprise governance; deals with major cloud providers funded purely by customer revenue.

How they differentiate

Unlike competitors that rely on fine-tuning open models (e.g., Fabraix), Abliteration.ai removes the refusal vector directly from weights and hosts it as a low-friction commercial API, reducing the compute and setup burden for teams that want guardrail-free frontier models.

Main competitors

Fabraix (agent red-teaming firm), Safe Intelligence, uncensored/abliterated model hosting platforms such as Featherless and Empero, the broader ecosystem of abliterated models on Hugging Face.

Key partnerships

Member of NVIDIA Inception and LAUNCH; collaborates with Microsoft for Startups (official homepage), Hosts abliterated open-weight models from Z.ai (GLM-5.3 / abliterated-model-large-v2; GLM-5.2 / abliterated-model-large), Deals with several major cloud providers funded by customer revenue (per co-founder Devon / TechCrunch)

Notable customers

Several early-stage red-teaming startups in the UK and Europe (including firms that red-team agents for banks, airlines, and critical-infrastructure operators).

Major milestones

Founded late 2025, officially incorporated March 2026, launched hosted abliterated GLM-5.2, then GLM-5.3 ("Abliterated Large v2"), struck deals with major cloud providers, began talks with VCs (as of Sept 2026).

Market positioning

Early-stage, bootstrapped startup operating at the controversial edge of AI safety — selling guardrail-free access to open-weight frontier models for defensive red-teaming while drawing safety-community warnings about potential misuse and minimal identity verification.

Geographic focus

US-hosted platform; customers primarily early-stage red-teaming startups in the UK and Europe.

Latest news about Abliteration.ai

Abliteration.ai commercializes stripped-guardrail access to Z.ai's GLM-5.3

More AI in Cybersecurity companies

Official website: https://abliteration.ai