Abliteration.ai

What is Abliteration.ai

Abliteration.ai is an LLM provider designed for professional teams whose legitimate work is often blocked by the refusal behavior of other LLM providers. It offers OpenAI- and Anthropic-compatible APIs that serve models with reduced refusal behavior through a technique called abliteration. This allows security researchers, red teams, trust and safety teams, and ML engineers to run sensitive queries, test AI applications for jailbreaks and prompt injection, classify content, and generate synthetic training data at scale.

How to use Abliteration.ai

1. Sign up for a free account to get an API key.
2. Change the base URL in your existing OpenAI or Anthropic SDK to https://api.abliteration.ai/v1 (for OpenAI) or the corresponding Anthropic endpoint.
3. Use your API key in place of your usual key. Your existing code continues to work with this one-line change.
4. Optionally, set up a policy in the Policy Gateway to enforce your own usage rules, quotas, and audit logging.
5. Start making requests to the /v1/chat/completions endpoint (OpenAI-compatible) or the Messages API (Anthropic-compatible).

Features of Abliteration.ai

• OpenAI- and Anthropic-compatible APIs: Drop-in replacement with a one-line base-URL change.
• Reduced refusal behavior: Models are abliterated to reduce refusals, enabling legitimate sensitive queries.
• Policy Gateway: Policy-as-code enforcement with outcomes: allow, refuse, rewrite, redact, or escalate. Includes per-project keys, quotas, shadow/canary rollouts, and audit logs.
• Zero data retention by default: Prompts and outputs are processed and discarded, never stored or used for training.
• Audit logging: Every decision is logged with a reason code, exportable to Splunk, Datadog, Elastic, or S3.
• Synthetic data generation: Generate labeled datasets (e.g., preference pairs, eval rows, classifier examples) in formats like OpenAI training JSONL.
• Free tier: Test the models before spending anything.
• Usage-based pricing: Pay per token with prepaid credits that never expire, and monthly plans starting at $20.

Use Cases of Abliteration.ai

• AI red teaming: Test AI applications for jailbreaks, prompt injection, and other security vulnerabilities.
• Cybersecurity: Run authorized penetration testing, draft PoCs for CVEs, and develop exploit code.
• Trust and safety: Classify content, generate edge-case examples, and build moderation systems.
• Synthetic data generation: Create training data, preference pairs, evaluation sets, and labeled datasets for ML pipelines.
• ML research: Conduct research that requires unrestricted model behavior.
• Defense and government contractor workflows: Support high-control environments with policy enforcement and audit trails.

Pricing

• Free tier: Available for testing.
• Builder Plan: $10/month.
• Team Plan: $25/month.
• Usage-based: Effective pricing is approximately $3 per 1M tokens (input + output). Prepaid credits are one-time and never expire; subscription credits reset monthly.
• Policy Gateway: Enterprise add-on billed monthly, layered on top of base token bundles.

FAQ

What is an uncensored LLM API? An uncensored LLM API lets developers use less-censored models without provider-side refusal filters. Enterprise teams use this for authorized red teaming, cybersecurity testing, trust and safety research, synthetic data generation, and ML evaluation while enforcing their own legal and policy controls.

What is the Policy Gateway? Policy Gateway is the enterprise AI Gateway for abliteration.ai. It provides a secure control plane for LLM apps with policy-as-code guardrails, quota management, PII redaction, and compliance logging.

Which enterprise teams use abliteration.ai? Enterprise security teams, AI red teams, cybersecurity labs, trust and safety teams, ML researchers, synthetic data teams, and government or defense-contractor programs use abliteration.ai when default provider refusals block authorized work and they need their own policy controls.

How does Policy Gateway enforce policy? Send requests to /policy/chat/completions with your policy_id, policy_user, and optional project ID. The gateway applies your rules, quotas, rollout settings, and returns decision metadata for audits.

Do I need a separate subscription for Policy Gateway? Yes. Policy Gateway is an enterprise add-on billed monthly. It layers on top of your base token bundles and includes policy-as-code, quotas, rollout controls, and audit logging.

What data does Policy Gateway store? Policy Gateway stores policy configuration and enforcement metadata (decision, reason code, policy ID, project/user tags) for audits. Prompt/output retention remains off by default.

Do you support training data generation? Yes. abliteration.ai can generate training data, preference pairs, evaluation sets, labeled datasets, and edge-case examples through the OpenAI-compatible API. Policy Gateway can add governance, scoped keys, quotas, and policy logs to dataset-generation workflows.

Is abliteration.ai OpenAI-compatible? Yes. abliteration.ai offers an OpenAI-compatible /v1/chat/completions endpoint with the same request and response structure, so most OpenAI clients work by changing the base URL and API key.

Do you retain prompts or chat logs? No prompt/output retention by default. Operational telemetry (token counts, timestamps, error codes) is retained for billing and reliability.

How does usage-based token pricing work? Billing uses total tokens (input + output), including image token equivalents. Effective pricing is ~$3 per 1M tokens; subscription credits reset monthly and prepaid credits are one-time.