Esta página está disponível em inglês. Sua preferência por português foi mantida para as páginas traduzidas.
Forward Intelligence / OpenDetector
Prioritize AI-generated text for review. Evaluate prompt injection and jailbreak signals before your models and agents act. Choose the model and operating policy your workflow needs.
Evaluation by request. Production access, served versions, coverage and commercial terms are confirmed for your deployment.
Specialist models / Detection
Choose AI-text review, prompt protection or both. Evaluation is by request. The pilot confirms the served version, coverage, account access and price.
Prioritize text for human review using AI-generation signals.
Human, AI-generated, edited and paraphrased text in your required languages and lengths.
A detection signal supports review. It does not prove authorship or justify an automatic adverse decision.
Evaluate prompt injection and jailbreak signals before content reaches a model or tool.
Direct attacks, indirect instructions, jailbreaks and benign content that resembles an attack.
Detection complements access controls, tool permissions and review. A low score does not establish that content is safe.
A measured path to access
Selected model: OpenDetector Injection. The request keeps this selection.
The pilot confirms coverage, operating policy, API contract, account permissions and quoted rates before production use. Initial estimates and requirements still being defined are welcome.
Define text volume, languages, lengths and source types. Include difficult cases and benign content, with an agreed way to label them.
Set acceptable false positives, missed detections, review effort and response time. Define what happens when coverage is incomplete or the detector is unavailable.
Pin the served version and measure the pilot. Confirm the API contract, account permissions, rates and rollout decision before production use.
The signal and the decision
A score is one input to your workflow. Your policy decides when to record a signal, request review, or stop an action.
AI-text detection and injection detection run independently. Generated text may be benign; human-written text may contain an attack. Evaluate each against its own labeled examples.
Short text, unsupported languages, incomplete extraction or an unavailable detector need an explicit fallback. Do not turn missing coverage into an approval.
Pair prompt protection with bounded tool permissions, authorized retrieval and human approval for consequential actions.
Define coverage and decisionsPlan a standalone integration that scores submitted text and returns structured signals. Use it with your application and model provider.
Scope checks at user input, retrieval and tool boundaries. Start in observation mode, then agree when to review or block.
Scope reserved capacity, region, data handling and version control for a defined enterprise workload.
The pilot measures supported inputs and response behavior. Scores, categories and document coverage depend on the agreed serving contract.