Abliteration.ai offers a hosted family of models whose refusal behavior has been reduced, delivered behind endpoints that mirror the OpenAI Chat Completions API and Anthropic Messages API. Because the interface is a drop-in match, teams can point their existing SDKs at a new base URL and keep their code, tooling, and agent workflows intact. The stated audience is professional: security researchers reproducing and validating vulnerability disclosures, red teams probing their own AI applications for jailbreaks and prompt injection, trust and safety functions that need to reason about harmful content, and ML engineers generating labelled training data at scale.
Beyond the models, Abliteration bundles a Policy Gateway that turns guardrails into something you write rather than something imposed on you. Policies are expressed as code and evaluated per request, so a workflow can allow a security task, refuse a conduct task, rewrite tone, or mask PII — each with a logged, auditable decision. A zero-data-retention option means prompts and responses are processed and discarded, and integrations can stream audit metadata into SIEM tooling such as Splunk, Datadog, Elastic, or Azure Monitor.
Commercial terms are usage-based. Monthly plans start at $20 (Developer, with matching usage credit), $50 (Growth), and $200 (Scale), each bundling credit and progressively larger API discounts; Enterprise adds SSO, SLAs, HIPAA/BAA, and region-specific hosting. Pay-as-you-go credit packs of $20, $100, and $500 are also offered, and the site indicates a free way to test the model before committing.