What AI Watchdog does
- Helps users navigate public records — campaign finance, congressional activity, judicial disclosures, federal spending — by surfacing source-linked summaries.
- Categorizes documents and signals using structured queries against primary-source datasets.
- Provides accountability research signals (e.g. donor-influence signals, contribution-to-policy correlation, accountability risk scans, elevated-review signals) as starting points for human investigation.
- Links every signal back to the underlying public record so users can verify independently.
What AI Watchdog does not do
- Does not make legal findings of corruption, fraud, quid pro quo, or wrongdoing.
- Does not establish causation from correlation between contributions and votes.
- Does not target private citizens. Subjects are public officials and entities with public records.
- Does not replace journalistic, legal, or academic review. Outputs are research aids.
Source verification
Every signal in AI Watchdog is intended to link to a primary public record (e.g. FEC filings, congressional disclosures, judicial disclosures, USAspending awards). Where a signal cannot link to a primary record, the signal is suppressed or labeled as non-source-linked. Users should always confirm against the underlying record before acting on a signal.
Human review
Research outputs intended for publication are reviewed by a human editor before they are framed as findings. Signals shown directly to users in the app are labeled as research signals, not legal conclusions. A human-in-the-loop review queue is part of VigilData's responsible-AI roadmap.
Bias & false-positive risk
AI-assisted categorization can over-fit on documented entities (who appear in more public records) and under-cover entities whose records are less digitized or less centralized. AI Watchdog signals reflect what public records permit us to document — they are not a complete picture of influence or wrongdoing. False positives are possible; signals should always be verified against primary sources.
Privacy & data use
AI Watchdog operates on public records about public officials and entities. It does not target private citizens. Where user-supplied text is sent to model providers, that data is processed for the specific query and not used to train external models. We minimize user-identifying information sent to model providers.
Model and provider disclosure
AI Watchdog uses third-party large language models for summarization, categorization, and source-trail capture. The specific model provider(s) in use will be disclosed in this section as the platform matures. Where available, the model name and version used for a specific output will be exposed alongside that output.
Rate limits and cost controls
Public AI Watchdog usage is rate-limited to keep the tool open and affordable. Heavy-use research partnerships are handled separately via research@vigildata.org.
Appeals and corrections
If a person or entity believes a signal misrepresents the underlying public record, they may submit a correction request to corrections@vigildata.org. VigilData's corrections policy is published at /corrections.
Language we avoid
VigilData does not describe AI Watchdog as a "corruption detector" or as producing legal findings. Internal labels such as donor-influence signal, contribution-to-policy correlation, accountability risk scan, and elevated-review signal are intentionally framed as research signals, not legal conclusions.
See /methodology for VigilData's broader methodology and /for-funders for funder materials on responsible-AI safeguards.