Trust and Safety

Fyreflock pairs adults with one stranger at a time, on video or on text, on the website and through an AI assistant such as Claude or ChatGPT. This page says who runs it, how the 18+ rule is enforced, what happens when you report someone, what we remove and report, and what reaches a model. Every line describes something the product does today; the bracketed lines are still to be completed, and say so.

Who operates it

Fyreflock is run by one independent developer in India, trading as Fyreflock. There is no company and no office. Operator of record: [Fyreflock to complete: legal operator line (trading name as registered, city and state)].

Every address below reaches that person, on IST working days. Support and law enforcement requests: [email protected], answered within 2 business days. Privacy, data access and deletion: [email protected], within 30 days. Complaints, illegal content and child safety: [email protected], acknowledged within 24 hours. The Contact page has the full table.

Who may use it

You must be 18 or older. One threshold for everything: holding an account, being matched on video or text, being in a Roost, and connecting an assistant.
You give your date of birth at sign-up. The server checks it and refuses to complete the account if you are under 18. There is no softer path.
The date is written once and you cannot change it afterwards. Age is worked out from it live, on every match and every assistant connection, so editing a stored flag gets nobody past it.
Someone who got in with a false date can be reported as Underage, a category of its own, and those reports go to the top of the queue. An account we conclude belongs to someone under 18 is banned, with the device it used, and the account and everything it created are deleted.

A date of birth is something a person typed. We do not verify identity, age or background beyond it, and the Terms say so. The camera check below is a check on the camera, not on the person.

The camera check

Every account passes a live-camera check once, at sign-up, and on video your camera must be on before you can enter the queue.
It runs in your browser and looks at pixels: a picture that is not blank and is changing over time. A covered lens, a lens cap, a photograph held up to the camera and most virtual-camera stills fail it.
On video, a face-presence model also runs on your device while you wait and during the call. No face in frame for 15 seconds pulls you out of the queue; 30 seconds on a call ends it, with a countdown you can see.
Nothing is stored. No frame of the check is uploaded, and it does not recognise a face, identify anyone or confirm an age. The only picture we ever hold from a call is the still frame attached to a report.
It is code running on your own device, so it stops the careless rather than the determined, and a recorded video of a real person would pass it.

Reporting

Every call and every chat on the website has a report button. Categories: nudity, underage, harassment, violence, scam, other, plus an optional note.
The evidence attaches itself: the text of that conversation (up to its last 200 lines), a still frame of the other person’s video on a call, and both account identifiers. You capture nothing, and the rule against recording others does not apply to the report button.
Text typed in a browser-to-browser conversation is never stored anywhere and dies with the call, so a report is the only way the words from one reach us.
When the conversation passed through our server (assistant mode), we attach our own copy of what was said, held in memory for up to 30 minutes after it ends. A transcript supplied by the reporter is not evidence and is not used.
Inside an assistant, reporting is a tool rather than a button: say “report them”. It takes the same categories, and the assistant cannot supply the transcript either. “Block them” works the same way.
A person reads every report. Underage reports sort first, then nudity, then the rest. Two different people reporting the same account moves it into a restricted pool, matched only with other reported accounts, until the reports are resolved.
Each report is closed as banned, warned or dismissed, by a person, and written to an audit log. A ban closes every other open report against that account. One report per conversation per reporter, and no more than ten an hour, so the button cannot be turned into a weapon.

Illegal content and child safety

Zero tolerance, for conduct as well as material. Our published Child Safety Standards cover it in full. In short:

Sharing, requesting or storing child sexual abuse material, grooming, sextortion, trafficking, or sexualising a child in any way ends the account permanently, with no warning, and bans the device it used.
We report suspected child sexual abuse material to the National Center for Missing & Exploited Children (NCMEC) and to law enforcement, and we preserve the evidence for them. An open report is never aged out.
Reporting runbook: [Fyreflock to complete: the NCMEC CyberTipline runbook: who files, within what time, what is preserved and for how long. Listed as a pre-launch item in the README and not yet written].
Other unlawful content is removed within 24 hours of a valid complaint to the grievance address, ahead of the full investigation.
We disclose information when the law compels us, and where we believe someone is at risk of serious harm.

If a child is in immediate danger, contact your local emergency services first. We are not a substitute for them.

Intimate images shared without consent

This is our notice under the Take It Down Act. Fyreflock has no feed, no gallery and no uploads, so an intimate image can only exist here as a live camera frame during a call or as the still frame attached to a report. If an intimate image of you, real or generated, was shown or shared on Fyreflock without your consent:

Email [email protected] with your name, or the name of the person acting for you; a way to reach you; where and when it happened on Fyreflock, with the username if you have it; and a statement that you are the person shown and did not consent.
We act within 48 hours of a valid request, and our own target is 24 hours: any copy we hold is deleted, the account that showed or shared it is banned, and we write back telling you what we did.
Nothing is kept except what the law requires us to preserve for a report to law enforcement.

Blocks and bans

Block ends the conversation at once. The other person is not told, and the two of you are never matched again, in either direction. Blocking does not file a report; do both if you want somebody looked at. Your blocked list is in Settings, where you can unblock.
Ban is decided by a person, with a reason, and audited. It ends every session and every connected assistant at once, bans the device, and dissolves any Roost the account owned.
A ban outlives the account. A banned account cannot reach the delete button, and deleting an account leaves a one-way hash of the email and device that records only whether you left with a ban or an open report, so signing up again comes back restricted rather than clean.
If you think a ban is wrong, write to [email protected]. It is read by a person, but the Terms do not promise an appeal.

What an assistant sees

On the website, video, audio and typed messages travel directly between the two devices, encrypted, and our server cannot read them. That changes only when one side is on an assistant:

When you talk through Claude, ChatGPT or Claude Code, or the chat box inside an assistant cannot open a direct connection, each line passes through our server in plain text. We hold those lines in memory for the conversation and 30 minutes after, for reports, and never write them to disk unless a report is filed. Never otherwise: a browser-to-browser chat is never routed through us.
They also pass through the AI provider you chose, or the one the other person chose, under that provider’s terms. A person using an assistant is shown as one to whoever they meet. In the text lobby you are matched with people who do unless you turn that off, with the button under Start.
Connecting an assistant asks you for a separate consent that says exactly this. It is versioned, and a change to it re-asks every connected assistant.
A stranger’s words reach your assistant wrapped and labelled as data written by another person, not instructions, with a fresh random nonce on every result so the wrapper cannot be forged. A line shaped like an order to your assistant is flagged with a visible warning rather than passed on quietly.
What your assistant sends for you is screened on our server before it leaves: anything shaped like a key, a token, a file path or a pasted document is refused, and so are contact details early in a conversation, with the assistant told why.
Banned accounts, guests and anyone under 18 are refused before a single tool is offered. Every connected assistant is listed in Settings, where you can disconnect it.

Use Fyreflock in a chat with no other connectors, files or tools attached. Whatever your assistant can see, a stranger can ask it about. The assistant guide explains the set-up.

UK Online Safety Act

Fyreflock is a user-to-user service that people in the UK can reach, so the Act’s illegal content duties apply to it, and the children’s safety duties apply unless under-18s are kept off it. What is in place: the service is 18+ with the gate enforced on the server; there is no feed, no recommendations, no public content and no way to reach a person except one conversation at a time; every conversation has a report button; and complaints are acknowledged within 24 hours. A typed date of birth is not highly effective age assurance in Ofcom’s terms, and we say so rather than claim otherwise.

Risk assessment: [Fyreflock to complete: the illegal content risk assessment and the children’s access assessment, the date each was completed, and the measures that followed].

India: DPDP Act and IT Rules

Fyreflock is operated from India and is an intermediary under the Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021. Complaints go to [email protected], acknowledged within 24 hours and resolved within 15 days; unlawful content and non-consensual imagery are removed within 24 hours. Grievance Officer: [Fyreflock to complete: name and designation of the Grievance Officer under Rule 3(2)].

Under the Digital Personal Data Protection Act, 2023, we process the data listed in the Privacy Policy for the purposes stated there, and none of it is sold. You can ask for access, correction or erasure at [email protected], answered within 30 days, and delete the account yourself from Settings at any time. Talking through an assistant is a separate purpose with its own consent. The Act treats anyone under 18 as a child; Fyreflock refuses under-18 accounts, so no child’s data is knowingly processed. Grievances about personal data go to the officer named above.

What we keep, and for how long

Calls and in-call text: never stored. The last 200 lines of a chat leave the call only inside a report.
Lines relayed for an assistant: in memory for the conversation and 30 minutes after, then gone, unless reported.
Friend messages: on your device, on a timer of 1 to 24 hours. A message waiting for an offline friend sits on our relay and is deleted on delivery or within 24 hours.
Who was matched with whom, and when: 90 days, then deleted by a sweep that runs on the server. Never the contents.
Reports: kept while open, however old. A resolved report and its chat log are deleted 365 days after it is closed.
Sessions, sign-in codes and assistant tokens: deleted by the same sweep once they expire or are revoked.
Deleting your account removes profile, friendships, messages, sessions and preferences immediately. What stays: reports others filed about you, payment records, and the one-way hash described above.

Contact

Safety, illegal content and complaints: [email protected]. Everything else: [email protected]. Reporting someone is faster in the app, because the evidence attaches itself. Put URGENT in the subject if a minor or a credible threat is involved.