CONTENT SECURITY
Security Policy
Where your content is handled and stored, and who can open it.
This policy covers studio, streamer and label work under contract.
Encrypted on your machine, before it leaves your network.
We give you a small program that seals (encrypts) every file to a Familiar Labs public key, the same family of cryptography that protects SSH logins. Only our private key opens it.
Stored and dubbed on our own servers in Toronto.
The sealed files go straight to the building where our team works: our own eight-figure cluster of NVIDIA B300 GPUs, with solid-state and network-attached storage on site.
WHERE YOUR CONTENT GOES
- Our own servers in Toronto, Canada
- Our own GPU cluster
- Our own storage, in the same building
WHERE IT NEVER GOES
- A public cloud, such as Google Cloud or AWS
- A third-party inference provider
Access by named officers.
Two custodians hold the decryption keys. One named researcher handles the content, and a different officer signs off its deletion.
Nobody outside these named roles can open your content: no other employee, contractor or vendor.
SECURITY LEAD
Mingi Kwon, Chief Technology Officer
INCIDENT LEAD
Michael Chan, Chief Business Officer
KEY CUSTODIAN
Mingi Kwon, Chief Technology Officer
SECOND KEY CUSTODIAN
Jibin Song, Chief Scientist
CONTENT HANDLER
Jaeseok Jeong, Principal Researcher
DELETION OFFICER
Mingi Kwon, Chief Technology Officer
ACCESS APPROVER
Andrew Liu, Chief Executive Officer
CHANGE APPROVER
Jibin Song, Chief Scientist
The finished dub comes back sealed.
Deleted on your terms, or after 14 days.
- ON REQUEST
Zero data retention
Deleted the moment the sealed dub is returned. - ON YOUR SIGN-OFF
Deletion on sign-off
Once you approve the dub, we delete your content immediately. - BY DEFAULT
14-day automatic deletion
With no instruction, a scheduled job deletes your content 14 days after delivery. The 14 days are there for your review and any changes.
We keep your content 14 days at most. Once deleted it is gone: there is no backup and no other copy, so it cannot be recovered. For a later change, you send the files again.
Your content is never used to train.
The decision recordTEXT ONLY
Written policies, named owners, reviewed every year.
CRITICAL
4 hours
HIGH
1 day
MEDIUM
3 days
LOW
7 days
Michael Chan, our Incident Lead, runs every response. If your content is affected, we tell you promptly and update you at least every 24 hours until the incident is closed.
PEOPLE
Background checks by the Toronto Police Service, signed confidentiality agreements and security training every year.
THE SITE
A locked server room: individual entry, cameras, an alarm, escorted visitors.
NETWORK
The content network is kept apart from the office and the internet. No remote access.
TESTING
An independent penetration test every year. Scans every month and every quarter.
CHANGES
Every change is approved by someone other than the person making it.
CONTINUITY
A written plan brings the service back after an outage. Your content is never backed up.
VENDORS
No service provider handles your content.
RISK
A security risk assessment every year, with each risk tracked to closure.
Every MPA control, answered.
The Motion Picture Association's Content Security Best Practices are the standard studios assess vendors against. Open a section for our answer to each control.
Familiar Labs is a member of the Trusted Partner Network+ (TPN+), the Motion Picture Association's content security program.
Organizational security
- Information security management
- A written information security policy, approved by our executives, reviewed every year and after any significant change. This page is its public summary.
- Acceptable use
- Nobody posts or shares anything about a client's unreleased content or project without that client's written consent. The same policy covers AI tools.
- Business continuity
- A continuity plan with a named owner covers loss of power, network, systems and premises, and includes incident response.
- Disaster recovery
- Company systems are backed up and encrypted, and only authorized staff can change or delete a backup. Client content is never backed up: once deleted, it cannot be recovered.
- Data and assets
- Client content is our highest classification. Its life cycle is fixed: sealed upload, dubbing, sealed return, deletion.
- Risk management
- A security risk assessment every year and after any significant change, covering workflows, systems and people, with risks ranked by severity and tracked to closure.
- Background screening
- Everyone with access to client work is screened before they start, within Canadian law. Each named officer has passed a background check by the Toronto Police Service.
- On-boarding
- A signed confidentiality agreement, policy sign-off and security training come before any access. Each grant is logged.
- Off-boarding
- Access is removed and equipment returned on the last day, logged with the date. Confidentiality obligations continue.
- Training and awareness
- Security training at hire and every year, tailored by role, with a log of who attended.
- Contracts with service providers
- No service provider handles client content. If a project ever needed one, you would be told first and the provider held to this policy in its contract.
- Incident response
- A written process with a named lead: detection, analysis, escalation, response, evidence, remediation and a root-cause review. Affected clients are told promptly; for an incident involving our model, within 24 hours of discovery.
- AI and machine learning security management
- Dubbing with our own model is the service you contract, so its use is approved in writing. The policy covers everything derived from your content: inputs, prompts, outputs, embeddings, caches, logs and telemetry. None of it trains a model, and no weights or adapters are tuned on it. Data sources are reviewed for origin and integrity before use.
Operational security
- ReceivingNOT APPLICABLE
- We receive digital files only, no physical client assets.
- PackagingNOT APPLICABLE
- Nothing physical is packaged.
- ShippingNOT APPLICABLE
- Nothing physical is shipped.
- Transport vehiclesNOT APPLICABLE
- Nothing physical is transported.
- Work from home and remote workers
- Everyone works on site in Toronto. Nobody works remotely, and nothing outside the site can reach the content network.
- Remote sites and locationsNOT APPLICABLE
- One site. We run no remote locations.
- Tracking
- Every file carries a unique identifier from upload to deletion. Each movement is logged with its time, and the logs are kept for at least a year.
- High-security titles
- Unreleased titles run under your working title or alias, with access limited to named staff under individual confidentiality agreements.
- Destruction
- Files are erased on your instruction, on your sign-off, or automatically after 14 days, inside the standard's 30-day guideline. The officer who signs off a deletion is never the person who handled the content. Every deletion is logged and confirmed in writing on request. Retired drives are wiped, then physically destroyed.
Physical security
- Entry and exit points
- The server room is locked and separated from the office and from other tenants. No screen showing content faces a window.
- Visitors
- Visitors sign a log, show government photo identification and are escorted. The logs are kept for a year.
- Electronic access control
- Entry to the server room is by individual electronic credential, granted by role and administered by a named officer.
- Access logging and monitoring
- Every entry is logged by the system, and the logs are kept for a year.
- Alarm system
- An alarm covers the server room and every entry point, armed whenever the site is unattended, with individual codes and alerts to named staff.
- Keys
- Physical keys to restricted areas are inventoried, signed in and out, and kept in a safe. A key that cannot be accounted for means new locks.
- Replication facilitiesNOT APPLICABLE
- We replicate no physical media.
- Camera system
- Cameras cover the server room and its entry points. Footage is kept securely for at least 90 days and can be viewed by authorized staff only.
- Environmental controls
- Temperature and humidity in the server room are held in range, with alerts when they leave it.
- Data centers and co-locationsNOT APPLICABLE
- Our servers sit on our own premises. We rent no data center or co-location space for client content.
- Cloud providersNOT APPLICABLE
- Client content never reaches a cloud provider.
Technical security
- Data input and output workflows
- Sealed files arrive on a dedicated transfer system, isolated from the production network, and are scanned for malware before ingest. Transfers are pulled from the more secure side, and nothing stays on the transfer system past 24 hours.
- Systems configuration
- Every machine is built to a hardened baseline: no default or guest accounts, no unneeded services, no local administrator rights, removable storage and wireless transfer blocked on content systems, screens locked after 15 minutes.
- Default accounts
- Default accounts are found, renamed or removed, and their passwords changed before a system goes into service.
- Endpoint protection
- Centrally managed anti-malware on workstations, servers and storage, updated and scanned on a schedule.
- Mobile devices
- Company devices are fully encrypted, centrally managed, auto-locking and remotely wipeable. Phones and personal devices stay out of the server room.
- Security information and event management
- Logs from firewalls, authentication, storage, transfer and key management are collected centrally in real time, time-synchronized, alerted on, reviewed by assigned staff and kept for a year.
- Authentication
- A unique account for each person. Multi-factor authentication by authenticator app or hardware key on every internet-facing and administrative system. Passphrases of 16 characters or more, with lockout after five failed attempts.
- Authorization
- Least privilege throughout. Roles for development and deployment are separate, and every account is reviewed each quarter.
- Identity access management
- Role-based access, managed by authorized staff. No shared accounts. Service accounts are documented and reviewed.
- Corporate email filtering
- Company email filters phishing, malware and executable attachments, and blocks content files from leaving by email.
- Web portals
- Any client-facing portal runs on HTTPS with a current certificate and TLS 1.2 or higher, on a dedicated server in a demilitarized zone, with each client's access kept apart.
- Shared security responsibility modelNOT APPLICABLE
- No cloud provider handles client content, so no responsibility for it is shared.
- Cloud misconfigurationsNOT APPLICABLE
- Client content sits in no cloud environment.
- Secure software development life cycle
- Code review on every release, dependency scanning, security testing on change, and no credentials in source code. Client data is never used in development or test without your approval.
- Security by design and privacy by design
- Data protection requirements are set at design time for every system that touches content.
- Code repositories
- Repositories are encrypted, version-controlled and behind the same authentication and access rules as every other system.
- Content transfer systems
- Content moves only through Sealed Upload and Sealed Return, on dedicated systems apart from production. Each transfer is approved, and files clear the transfer system within 24 hours.
- Application configuration, licensed
- Licensed software is properly licensed and configured to its vendor's security guidance, reviewed each year: minimal administrators, unused features off, sessions closed after 15 minutes idle.
- Application configuration, developed in-house
- Our own tools follow written configuration guidelines, reviewed each year, with the same limits on privileged accounts, features and idle sessions.
- Network configuration
- Internet-facing servers sit in a demilitarized zone, one service each, with no route from there into the internal network. We keep an inventory of every exposed address.
- Point-to-point connections
- Any private connection is encrypted with AES-256 over TLS 1.2 or higher, and documented.
- Network topology diagram
- A current network diagram shows every segment, the internet boundary and the path client content takes. Your security team can see it under a non-disclosure agreement.
- Network traffic
- Managed switches with port security. Unused ports are disabled.
- Firewall access control list
- Stateful firewalls deny all traffic by default, in both directions. Only named protocols to named hosts are allowed, and rules and logs are reviewed regularly.
- Content and production networks
- The content network is segmented from the office network, the transfer zone and the internet. No machine bridges two networks.
- Firewall management
- Firewall administration is least-privilege, under change control and unreachable from the internet.
- Intrusion detection and prevention
- Network intrusion detection and prevention alerts on and blocks suspicious activity, with signatures kept current and all activity logged.
- Internet access
- Systems that store or process content have no general internet access.
- Remote access
- No remote access to the content network. All staff work on site in Toronto.
- Web filtering
- Peer-to-peer sharing, malware and known malicious sites are blocked.
- Wireless networks
- No wireless on the content network. Office wireless uses WPA2-Enterprise or WPA3, and the guest network reaches the internet only.
- Cloud service provider and consumerNOT APPLICABLE
- We run no cloud tenancy for client content.
- Network connections
- Connections are encrypted, limited to authenticated and authorized systems, and monitored. An unauthorized connection is detected and removed.
- Encryption
- AES-256 for content at rest, and AES-256 with TLS 1.2 or higher in motion, on top of the file-level sealing. Drives and backups are encrypted.
- Key delivery messages and trusted devicesNOT APPLICABLE
- We do not create or handle digital cinema packages.
- Encryption key management
- Decryption keys never travel with the content and are never stored beside it. Two named custodians hold them, our Chief Technology Officer and our Chief Scientist, apart from the researcher who works on the content, and they are rotated on a set schedule.
- Vulnerability management
- External addresses are scanned every month and internal systems every quarter, and again after any significant change. Every finding gets a remediation plan.
- Penetration testing
- An independent, certified tester runs a penetration test every year and after any significant change, covering the network, the transfer tools and our interfaces.
- Patching
- Security patches are tracked from vendor advisories and applied on a schedule. Unsupported systems are retired.
- Change control
- Changes to systems, networks and software are reviewed, approved and recorded against a current inventory, each with a tested way back. The approver is never the person making the change.
- Authentication
- Every user, system and service authenticates before it joins a workflow. Connections between services are mutually authenticated and encrypted.
- Authorization
- Every connection is allowed by a written authorization policy.
- Protect surfaces
- Client content is the protected surface. Each point of access to it carries its own check.
- Ingress and egress transfer
- Your content is never used to train or fine-tune a model; that would take your written consent, and we do not ask for it. Models, datasets and libraries from public sources are scanned and brought in through a segmented, restricted network before use.
- Application configuration
- Our model runs on our own hosts, configured to written guidelines reviewed each year, with encryption and access control at the application level.
- Governance
- We keep an inventory of approved AI tools, review it each quarter, and nothing outside it touches client content. Your written approval is recorded per project, and on request we tell you which AI tools touched your content. Every action of our model on your content is logged with the account, the time and the output. Prompts and configuration are classified as high as the content they touch. Behavior outside the defined task is detected and treated as an incident, reported to you within 24 hours of discovery. AI governance is part of on-boarding and yearly training.
- In-house development
- Each client's content, session data, prompts and outputs are kept apart, with no shared context, cache or knowledge base between clients. The decision record belongs to one client. Session data is deleted at the end of the project, on the same schedule as the content. Every AI component is inventoried with its version, source and approval, and verified for integrity before use. Code review, security testing and change control extend to model configuration. Scripts, subtitles and metadata you send are validated so that instructions embedded in them cannot steer the model. For high-security titles, access to logs, prompts and outputs is restricted to named staff.
- Third-party AI services
- No third-party AI service receives client content. Any exception would need your written approval, with its scope and duration on record, zero retention, the provider's commitment not to train on it, and proof that it keeps its clients apart.
- Agentic capabilities
- No autonomous AI agent acts on client content. A person approves anything that moves or deletes content outside the schedule you set. If an agent were ever introduced, it would carry its own identity, least-privilege access to one project, an allowlist of tools and a way to suspend it at once.
- Testing
- Our model's interfaces are inside the yearly penetration test: the separation between clients, inputs crafted to misuse the model, and the published top ten risks for language-model applications. AI components and dependencies are scanned for known vulnerabilities before they are integrated.
- Error handling and logging
- Error messages shown to users reveal nothing about our systems. Application errors are logged centrally.
- Session management
- A new, cryptographically random session token at each sign-in, kept in secure cookies and out of every URL. Sessions end after 15 minutes idle.
- Application hosting
- HTTPS everywhere, with TLS 1.2 or higher and strict transport security. Application, database and storage sit in separate network segments, with unused services and ports removed.
- Application security testing
- Our own applications are tested each year with static, dynamic and fuzz testing, and the results are kept on file.
▚ At a glance ▞
WHERE
Toronto, Canada
WHO HOLDS THE KEYS
Two named custodians
IN TRANSIT
Sealed on your machine
RETENTION
Zero on request; 14 days by default
TRAINING
Never
PUBLIC CLOUD
Never
We answer your security questionnaire and sign a non-disclosure agreement before any content moves.
Email the team