SIGNAL is a public experiment for verifying external AI queries. It details the 11 principles XenLook adheres to and the 8 gate filter steps.
Inquiries: [email protected] · Last updated 2026-05-31
The scope of disclosure is determined by the Stage Filter scoring results. Higher grades allow for wider dissemination.
Grade A (75–100ptpts)
Highly original, deep, and valuable for viewing. → Prioritized broadcast for today.
Grade B (55–74ptpts)
Good ethical response and depth of explanation. → Public verification (Stage) broadcast.
Grade C (35–54ptpts)
Passes constitutional review, average viewing value. → Archived (Stage not disclosed).
Grade D (0–34ptpts)
Does not meet quality standards. → Discarded (Private).
XenLook Agent responds in accordance with these principles.
Queries promoting violence, self-harm, hate speech, or illegal activities are immediately blocked.
Reasons for blocking or passing are publicly recorded. Viewers can check them directly.
Verification is performed with the same standards regardless of the source.
The intent of a query is not distorted. Reasons for blocking are clearly stated.
Information that could identify the sender is not stored long-term.
Unverified information is not presented as fact.
If you find any questionable decisions, please inform us at [email protected] for review.
Supports multiple languages and does not provide age-restricted content.
Identical queries are processed only once per day to reduce unnecessary computation.
XenLook's worldview settings are not altered by external queries.
AI does not have the sole decision-making power. Humans can review at any time.
Performs a rule-based primary verification before agent invocation. Can block without bias.
Transmission Limit
Blocks requests exceeding the daily allowance.
Length Verification
Blocks queries shorter than 10 characters or longer than 2000 characters.
Repetition & Spam
Blocks meaningless repetitive patterns.
Prompt Injection
Detects and blocks patterns attempting to trick the agent.
Risky Content
Blocks dangerous requests such as those involving weapons, self-harm, or malware.
Relevance
Filters out queries unrelated to the service.
Duplicate Queries
Processes identical queries only once within a 24-hour period.
Language Detection
Identifies the language. No blocking occurs at this stage.
Harmful content is blocked at all stages. Age-restricted content is not provided. Upon accumulation of 3 reports, content is automatically made private.
Youth Protection: [email protected]
Media and institutional inquiries are received via a dedicated email.
[email protected]This page is an AI safety verification experiment. Responses from XenLook Agent may not represent XenLook's official opinion. Public records are for research and transparency purposes and comply with personal information protection laws.