Trust & SafetySafety is not a feature. It is the foundation.
Candidfy was built for honest communication. That mission fails completely if the platform becomes a vehicle for harm. This page describes how Candidfy protects recipients, protects senders, and makes those two things compatible.
The four-stage safety pipeline
Every message sent through Candidfy passes through four stages before the recipient ever sees it:
01
Intent classification
The raw message is scored across harm categories including explicit threats, harassment, coercion, stalking language, and sexual coercion. Messages above the threshold are blocked immediately and never proceed. When blocked, the sender is told exactly why and offered specific reframe suggestions — not left at a dead end.
02
AI rewriting
Messages that pass intent classification are rewritten by AI — calibrated by relationship type and tone target chosen by the sender. The rewrite preserves every honest observation and removes every expression of cruelty, aggression, or shaming. The core truth is always kept intact.
03
Sender approval
Nothing is delivered without the sender explicitly approving the rewritten message. This creates a natural pause — a moment of reflection between raw feeling and actual delivery. Senders can edit the rewrite, start over, or abandon the message entirely.
04
Recipient controls
Recipients can report any message they receive. Reports go directly to a human review queue. Senders who generate reports face escalating restrictions. Opt-out is permanent and enforced at the platform level — any sender attempting to contact an opted-out recipient is blocked regardless of message content.
What Candidfy will never deliver
Regardless of how a message is framed, these are hard blocks that no rewriting can circumvent:
✕Explicit threats of physical harm or death
✕Stalking language — references to monitoring location, schedule, or movements
✕Sexual coercion or non-consensual intimate content
✕Messages to recipients who have opted out of the platform
✕Coordinated harassment — multiple messages to the same recipient in a short window
✕Content targeting minors
The ML learning system
Candidfy's safety system learns continuously from real usage. Every blocked message, every recipient report, and every sender false-positive flag contributes to a labeled dataset that improves the classifier over time. A human reviews every flagged item before it influences the model — there is no fully automated retraining.
This means the system gets more accurate as it sees more messages — better at catching genuine harm, better at passing legitimate emotional content that static rules would wrongly block.
Recipient protections
Permanent opt-out
Any recipient can opt out permanently at candidfy.com/optout. This is enforced at the platform level — not just a preference. Any sender attempting to reach an opted-out contact is blocked regardless of message content.
One-click reporting
Every message includes a report option. Reports go to a human review queue sorted by priority. Recipient reports are the highest-priority signal in the system.
Rate limiting
No sender can send more than three messages to the same recipient within a 30-day window. This limit is enforced by the platform and cannot be circumvented.
No reply required
Recipients are never pressured to respond. The reply option exists but there is no notification, no nudge, and no follow-up if they choose not to engage.
Sender accountability
Senders are never anonymous to Candidfy itself — only to recipients. Every message is linked to a sender token. If a recipient reports a message and a human reviewer confirms harm, the sender token is flagged and restrictions escalate with each confirmed violation. Repeat violators are permanently blocked.
In cases involving credible threats of physical harm, Candidfy will cooperate with law enforcement. Anonymity is a feature for honest communication — not a shield for harm.
Contact & reporting
If you received a message that made you feel unsafe, or want to report abuse of the platform:
Candidfy operates a zero-tolerance policy on platform abuse. The same anonymity that protects senders who need courage also requires a robust system to prevent misuse. These two things are not in tension — they are both necessary for the platform to work.