Privacy and evidence use — version 2

Operator: Mike Spicer (d4rkm4tter). Administrative deletion contact: michael.spicer@corelight.com.

This site accepts cybersecurity AI prompts and supporting evidence for research into model refusal behavior and for creating reviewed, reproducible benchmark candidates. A contribution without an account is an observation, not proof that a model refused or that the refusal was incorrect. No account, name, or email is requested. Your source IP address and submission request headers are retained with the evidence and may identify you. Do not include passwords, API keys, personal information, or confidential material. Conversations, HAR files, logs, agent traces, and HTTP headers may contain these inadvertently.

The public intake service receives submitted text and attachments. It returns only an opaque local receipt and cannot retrieve submissions or query the private analysis environment. Accepted files are quarantined and made available to restricted private researchers/reviewers. Private workers may extract text from supported formats without following links or running uploaded code. Research staff can classify, redact, and create reviewed benchmark cases. Raw submissions will not be publicly published without separate permission. Redistribution of reviewed material also requires a separate permission grant and independent release approval.

We do not use advertising trackers, browser session replay, or contributor accounts. Internet/network providers and required AWS infrastructure may see source IPs and request metadata. For accepted submissions, we retain the trusted client IP (as supplied by the load balancer), the edge peer IP, request time, host, HTTP protocol, and all submission request header names and values delivered to our application. Forwarded headers can contain unverified claims. Headers can include user-agent strings, referers, cookies, or authorization values if your client sends them. These details stay in restricted original evidence for analysis; they are not included in benchmark exports. Application rate limiting also briefly uses an in-memory pseudonymous IP-derived key. The operator does not promise anonymity. Application operational logs are configured without submission contents and kept for up to 14 days; audit trails for private reviewer actions are kept for up to 90 days. Cloud-account security logging, where required, can independently retain service metadata.

Local unpublished intake copies, including request metadata, are retried for up to 7 days. Landing copies are scheduled to expire after 14 days. Restricted originals and unredacted private analysis copies, including request metadata, are scheduled to expire after 90 days. Reviewed/redacted research artifacts are scheduled to expire after 365 days unless their retention is renewed for a recorded research purpose. Encrypted recovery backups expire after 14 days. Cloud lifecycle deletion can lag its scheduled date; operational checks investigate overdue deletion. Exported dataset releases cannot necessarily be recalled from previous recipients.

There is no public status, download, or deletion endpoint. To request deletion without an account, email michael.spicer@corelight.com with your receipt ID. The operator may accept the receipt as the deletion claim; anyone holding your receipt could request deletion, so keep it private. Deletion is handled administratively in the private environment. Landing and any local intake copies expire by their fixed schedule rather than via a callback; backup copies expire on their recovery schedule. A private deletion register prevents routine reimport of a removed record. Do not email evidence or secrets unless specifically requested through an approved private process.

Questions about access, use, deletion, or publication permissions can be sent to michael.spicer@corelight.com.