# Mothlight

> Mothlight is an open observatory of bots and AI agents attacking the web: a multi-region honeypot network that labels every automated visitor, scores how likely it is to be AI-driven, and publishes the data.

Where bots come to be seen.

Data mode: Observed sensor data.
Snapshot: 2026-11-02T00:34:23.267Z. Window: rolling last 24 hours.
42,485 automated requests; 0.1% in AI-related categories (AIC, AIA, AIX).
This is sensor traffic, not all internet traffic. AI category share does not prove every request used an LLM or was malicious.

## Research and evidence
- [Live observations](https://mothlight.ai/live): recent requests and provisional labels
- [Bot directory](https://mothlight.ai/bots): stable IDs, confidence and identity verification
- [Reports](https://mothlight.ai/reports): daily statistical digests with UTC windows
- [Methodology](https://mothlight.ai/methodology): labeling, confidence, AI likelihood and limitations
- [Robots compliance](https://mothlight.ai/robots-compliance): controlled rule tests and observed violations
- [Canaries](https://mothlight.ai/canaries): behavioral evidence, without active token values
- [Threats](https://mothlight.ai/threats): observed attack patterns and defensive context

## Reuse
- [Dataset](https://mothlight.ai/data): samples and aggregates, CC BY 4.0 (samples require an approved read key)
- [API reference](https://mothlight.ai/api-docs): aggregates without a key; event-level and bulk endpoints need an approved key
- [Overview JSON](https://mothlight.ai/api/v1/stats/overview)
- [RSS](https://mothlight.ai/feed.xml)
- [Extended machine-readable guide](https://mothlight.ai/llms-full.txt)
CC BY 4.0 applies strictly to public samples and aggregates: https://creativecommons.org/licenses/by/4.0/
Attribution for those public releases: Mothlight, source URL, UTC observation window, access date, CC BY 4.0. Indicate modifications.
Gated event data, raw data, realtime feeds and training exports are proprietary. Access does not grant redistribution or training rights; request terms at help@mothlight.ai.

## Access matrix
- Anonymous: aggregates that mirror the public website only (stats, regions, reports, robots-compliance aggregates, known-bots, bot directory and profiles). Event lists, event details, /stream, /export, /canaries and /prompts return 401 with the request-form URL.
- read: approved key (issued only after a human approves a request at /api-docs#request-api-access): aggregates, CC BY 4.0 export samples and delayed event teasers, with higher request quotas. Event lists allow 50 rows/page and 200 rows/window, with a maximum 24-hour window ending at least 60 minutes ago. A read key does not unlock gated data.
- events: gated event lists and details, including sensitive metadata and IP filtering. No per-event SSE or full export without the corresponding scope.
- stream: gated event access and per-event SSE. Other callers receive aggregate SSE only.
- export: gated event access and full JSON, CSV or NDJSON export, up to 100,000 rows/call.
- raw: all gated capabilities, raw IPs and payload/training JSONL export. A query parameter never grants a scope.
- Keyed export: /api/v1/export?format=ndjson with an approved read key returns a delayed, suppressed CC BY 4.0 sample of at most 1,000 rows. Without a key it returns 401. Full export requires export or raw; format=jsonl requires raw.
Keyed export samples include deterministic synthetic watermark rows. These are provenance markers, not observed attacks; operators detect them using ML_WATERMARK_SEED and the sample UTC day.

## Ethics
- [Ethics and disclosure](https://mothlight.ai/ethics)
- [Security contact](https://mothlight.ai/.well-known/security.txt)
- [About Mothlight](https://mothlight.ai/about): a SEWORKS observatory
- [Contact and licensing](https://mothlight.ai/contact)
Passive inbound observation only. No active scans, visitor callbacks, code execution, or real-model calls from fake endpoints.
Brand-site robots.txt allows every crawler, including AI crawlers. Decoy experiments are separate.
Contact: help@mothlight.ai.
Language: English.