-
Notifications
You must be signed in to change notification settings - Fork 0
Legal Framework
ARGUS is designed from the ground up with legal defensibility in mind. This page explains the framework, the language choices, and the protections built into every published page.
Opinion vs. Fact
Courts distinguish between:
- Statements of fact — assertable as true or false → defamation exposure
- Statements of opinion based on disclosed methodology → protected expression
"John Smith is a fake account" = statement of fact = defamation risk.
"Our scoring system flagged this profile at 18/100 based on X, Y, Z signals — you decide" = disclosed algorithmic opinion = protected.
ARGUS is designed entirely around the second framing. We never assert. We analyze, score, and disclose. The visitor decides.
Every published page includes all five:
The full scoring algorithm is open source on GitHub. Anyone can audit exactly how every score is calculated. This is the strongest protection against claims of arbitrary or malicious targeting. If your methodology is public and consistent, it's very hard to argue you acted with malice toward any specific individual.
Every page displays the confidence percentage and estimated false positive rate at that confidence level. This demonstrates good faith — we're not claiming certainty, we're presenting probabilistic analysis with disclosed uncertainty.
Displayed prominently on every page, above the content. The exact language matters:
"This is algorithmic analysis, not an editorial verdict. Scores reflect automated pattern matching against known synthetic account signals. This analysis expresses an opinion based on disclosed methodology. You decide."
The dispute form is above the fold — not buried in a footer. This demonstrates good faith and willingness to correct errors. Platforms including Facebook and Twitter require evidence to dispute disabled accounts. ARGUS follows the same model.
Every page that goes live was reviewed and approved by a human editor (you). This shifts ARGUS from "automated publisher" to "editorial platform," which significantly strengthens legal protections.
These rules apply to all user-facing strings in ARGUS. Never use the prohibited words in any page, email, badge, or tooltip.
| ❌ Never Use | ✅ Use Instead |
|---|---|
| fake | flagged, high risk, suspicious |
| fraud | coordinated inauthentic behavior |
| criminal | policy violation (if applicable) |
| liar | content inconsistency detected |
| bot | synthetic account signals |
| scammer | high-risk engagement patterns |
Can publish:
- Algorithmic trust scores with full signal breakdown
- Behavioral pattern observations (account age, activity rate, posting patterns)
- Network graph observations (connection age distribution, coordination signals)
- Image analysis results framed as probability, not certainty
- Text analysis results framed as AI probability, not certainty
- Community report counts (how many people flagged, not who)
Cannot publish:
- Claimed criminal or fraudulent intent (requires evidence you don't have)
- Private information not publicly accessible on the platform
- Information obtained through unauthorized access to platform data
- Defamatory statements about specific named individuals
- Recommendations to avoid, block, or report an individual (present analysis, not recommendations)
The dispute system isn't just fair — it's a key part of your legal defense.
When a dispute is submitted and resolved (either way), ARGUS logs:
- Date received
- Evidence tier provided
- Decision made
- Resolution note
If you ever face a legal challenge, you can demonstrate:
- The analysis was algorithmic, not targeted
- The subject had the opportunity to dispute
- Disputes were reviewed in good faith
- The process was consistent across all subjects
A resolved dispute that keeps the page live is actually stronger legally than never having received a dispute — it shows the process was followed and the evidence was insufficient to meet the threshold.
If a European user requests removal of their profile analysis under GDPR's right to erasure:
- They submit a dispute with their identity verified (Tier 1 evidence)
- The page converts to a "dispute resolved" notice (no personal data, just the outcome)
- The underlying profile record in D1 is anonymized (handle replaced with a hash)
- R2 evidence is retained for 7 years for legal defense purposes
CCPA (California) follows the same workflow. Users in California have the right to opt-out of "sale" of personal data — ARGUS does not sell data, so this primarily concerns deletion requests.
Section 230 of the Communications Decency Act provides immunity to platforms for third-party content. ARGUS has elements of both publisher (you write the algorithm) and platform (community submits profiles).
The human approval step is important here: you are not blindly publishing algorithmic output — you are making an editorial decision on each piece. This makes the Section 230 picture cleaner than a fully automated system.
The following organizations publish similar analysis at scale and have established legal precedent for this type of work:
- Stanford Internet Observatory — publishes network takedown reports naming accounts
- EU DisinfoLab — identifies coordinated inauthentic networks publicly
- The New York Times — publishes investigations into fake account networks
- DFRLab (Atlantic Council) — documents influence operations with named accounts
Common thread: they show their work, disclose methodology, frame findings as analysis, and provide a mechanism for correction. ARGUS follows the same framework.
This page is not legal advice. If ARGUS faces a specific legal challenge, consult a qualified attorney. This framework represents the architectural decisions made to minimize legal risk, not a guarantee of legal immunity.