Privacy Policy

1. Introduction

San Francisco Labs Inc. ("San Francisco Labs," "we," "us," or "our") is dedicated to the safe, responsible, and rigorous advancement of frontier artificial intelligence. This Privacy Policy details our comprehensive approach to the collection, use, retention, and safeguarding of information across our ecosystem.

This policy applies to all individuals ("you") who interact with our website, research interfaces, computational APIs, open-source repositories, and proprietary model environments (collectively, the "Services"). By accessing our Services, you acknowledge that you have read and understood the data practices outlined in this document.

2. Information We Collect

Our data collection practices are strictly governed by the principle of data minimization. We only collect information that is technically necessary to provision our Services, secure our infrastructure, or advance our foundational research.

2.1 Information You Provide Directly

Account and Identity Data: When establishing access to our restricted-use APIs or research clusters, we collect identifying information such as your name, organizational or institutional affiliation, and professional credentials required for vetting.
Communications: Content submitted through our contact forms, research applications, vulnerability reports, or other direct communication channels.
Research Inputs: Prompts, queries, datasets, or fine-tuning materials explicitly uploaded by approved partners for processing through our frontier models.

2.2 Information Collected Automatically

Telemetry and Telemetry Logs: Time-stamped logs of API requests, inference latency, token usage, and payload sizes.
Network and Device Data: IP addresses, user-agent strings, request headers, and device metadata utilized strictly for rate-limiting, abuse prevention, and load balancing.
Security Analytics: Anomalous access patterns, failed authentication attempts, and cryptographic handshakes monitored by our cloud perimeter defenses.

3. How We Use Your Information

San Francisco Labs utilizes the collected data to execute our primary directive: developing safely scaled intelligence. Specifically, we process your information to:

Provision Core Services: To authenticate your identity, execute inference requests against our models, and route computational outputs back to you.
Ensure Infrastructure Security: To detect, prevent, and actively mitigate adversarial actions, including denial-of-service attacks, model extraction attempts, or unauthorized environment access.
Improve Foundational Models: Aggregated, anonymized telemetry may be utilized to identify systemic failure modes, optimize hardware utilization, and refine our alignment and safety guardrails. We do not use your proprietary inputs to train our foundational models without explicit, written agreement.
Legal Compliance: To adhere to applicable statutory obligations, export control laws, and enforceable regulatory requests.

4. Cloud Infrastructure and Data Residency

The entirety of our computational pipeline—from public-facing ingress points to deep, isolated training clusters—is engineered on Amazon Web Services (AWS).

By interacting with our Services, you expressly consent to the transfer, storage, and processing of your data within our highly restricted AWS Virtual Private Clouds (VPCs). We leverage AWS's shared responsibility model to ensure that data residency and physical security standards meet or exceed the highest industry benchmarks for frontier AI laboratories. All sensitive data is cryptographically secured at rest using AWS Key Management Service (KMS) and in transit via TLS 1.3.

5. Data Sharing and Disclosure

San Francisco Labs does not, under any circumstances, sell your personal information. Our data sharing is strictly limited to the following operational necessities:

Cloud Service Providers: AWS, as our foundational infrastructure provider, processes data strictly in accordance with our configured environments and their enterprise service agreements.
Legal and Regulatory Obligations: We may disclose information if legally mandated by a court of competent jurisdiction, or to comply with national security and export control regulations governing frontier AI technologies.
Protection of Vital Assets: We reserve the right to share specific network telemetry with cybersecurity partners or law enforcement if we detect a credible, imminent threat to our model weights, internal infrastructure, or physical safety.
Corporate Restructuring: In the event of an acquisition, merger, or asset transition, data access will be securely transferred under the strictest confidentiality agreements.

6. Data Retention Policies

We retain personal data only for the duration necessary to fulfill the purposes outlined in this policy. Inference telemetry and API logs are subject to automated rolling deletion schedules, typically anonymized or purged within 30 to 90 days, unless preserved for active security incident investigations. Research inputs submitted by institutional partners are governed by the specific retention schedules negotiated in their respective data processing addendums.

7. Your Rights and Jurisdictional Controls

Depending on your geographic jurisdiction (such as the GDPR in the European Union or the CCPA/CPRA in California), you possess specific, enforceable rights regarding your personal data. These may include:

The right to request a cryptographic export of the data we hold concerning you.
The right to demand the immediate deletion of your personal data from our active databases.
The right to request correction of inaccurate institutional or personal records.
The right to object to or restrict specific processing methodologies.

To exercise these rights, please utilize the Contact link provided in the footer of this website. Our compliance engineering team evaluates all verifiable requests within legally mandated timeframes.

8. Policy Modifications

The landscape of artificial intelligence and its associated legal frameworks is rapidly evolving. We reserve the right to autonomously update this Privacy Policy to reflect advancements in our security posture, shifts in our infrastructure architecture, or new statutory requirements. Substantive modifications will be immediately published to this endpoint, and your continued use of the Services constitutes acceptance of the revised protocol.

9. Foundational Security Philosophy

At San Francisco Labs, we operate under the premise that frontier artificial intelligence models are dual-use assets of immense strategic value. Consequently, securing our infrastructure, protecting our model weights from exfiltration, and ensuring the integrity of our training data are not secondary concerns—they are foundational to our existence.

We implement a defense-in-depth, "zero trust" architecture designed to protect against advanced persistent threats (APTs), insider threats, and sophisticated cyber-espionage. Our security posture is continuously audited, aggressively updated, and engineered to scale alongside our computational capabilities.

10. Cloud Infrastructure & AWS Architecture

Our entire computational backbone—spanning massive parallel training clusters, evaluation sandboxes, and high-throughput inference APIs—is hosted natively on Amazon Web Services (AWS). We leverage AWS's shared responsibility model, building our defenses on top of their compliance-certified physical infrastructure.

10.1 Network Isolation and Topology

Air-Gapped Training Environments: Our core training clusters operate within strictly isolated AWS Virtual Private Clouds (VPCs). These environments lack an Internet Gateway (IGW); they cannot initiate or receive connections to the public internet. Access is brokered exclusively through hardened, audited bastion hosts.
Inference Separation: Public-facing API endpoints operate in logically and physically separate AWS subnets from our internal research and training data lakes. We employ AWS Transit Gateways and rigorous security group configurations to prevent lateral movement.
DDoS Mitigation: Public ingestion points are shielded by AWS Shield Advanced and AWS WAF, providing automated, heuristic-based mitigation against volumetric and state-exhaustion DDoS attacks.

10.2 Identity, Authentication, and Access

Zero Trust IAM: All internal access is governed by the Principle of Least Privilege (PoLP). No employee has standing, persistent access to production databases or model weights.
Mandatory Hardware MFA: Access to our AWS control plane, code repositories, and deployment pipelines requires cryptographic, hardware-based Multi-Factor Authentication (e.g., FIDO2/WebAuthn security keys). Phishable forms of MFA (like SMS) are strictly prohibited.
Just-In-Time (JIT) Access: Temporary elevation of privileges requires peer-reviewed, documented justification and automatically expires after a predefined cryptographic window.

11. Model Weight Protection Protocols

The proprietary weights of our frontier models are our most critical assets. Protecting them from unauthorized exfiltration or copy-cat distillation is paramount.

Data-at-Rest Encryption: All weights are encrypted at rest using AWS Key Management Service (KMS) with Customer Managed Keys (CMKs) rotated automatically on a strict schedule.
In-Memory Protections: Where technically feasible on modern GPU accelerators, we utilize secure enclaves and confidential computing architectures to protect weights while loaded in VRAM during inference.
Egress Controls: Instances with access to model weights are bound by absolute egress firewalls. They cannot communicate with unauthorized S3 buckets, external IP addresses, or unverified endpoints. Any attempt to export data beyond a threshold triggers an immediate, automated network quarantine of the offending instance.

12. Application and Data Security

Beyond infrastructure, our application layer and data handling pipelines are subjected to rigorous defensive engineering.

Automated Vulnerability Scanning: All commits to our codebase undergo automated Static Application Security Testing (SAST), dependency scanning for known CVEs, and secret-detection algorithms before they can be merged.
Data in Transit: All data exchanged between San Francisco Labs and our users, or between our internal microservices, is encrypted in transit using TLS 1.3 with Perfect Forward Secrecy. We do not support legacy cryptographic suites.
Customer Data Segregation: For enterprise and research partners, logical segregation is enforced at the database layer. Prompts and inputs are never intermingled, and tenant-specific encryption keys can be provisioned upon request.

13. Continuous Monitoring and Threat Detection

We assume breach and engineer our monitoring systems to detect anomalies instantly.

Our Security Operations Center (SOC) aggregates logs from AWS CloudTrail, VPC Flow Logs, application telemetry, and endpoint detection agents into a centralized Security Information and Event Management (SIEM) system. We utilize specialized machine learning algorithms to baseline normal developer and API behavior, triggering high-severity alerts for deviations such as impossible travel, unusual query patterns, or unexpected data volume movements.

14. Incident Response and Recovery

In the event of a suspected security event, San Francisco Labs executes a pre-rehearsed Incident Response Plan (IRP).

Containment: Automated runbooks are designed to instantly sever network access, revoke IAM credentials, and snapshot volatile memory of compromised instances.
Eradication and Recovery: Infrastructure is rebuilt immutably from known-good, scanned codebases (Infrastructure as Code) to ensure no persistent footholds remain.
Transparency: Should an incident result in the unauthorized access of user data, we are legally and ethically bound to notify affected parties and relevant regulatory bodies in accordance with global data protection laws.

15. Personnel Security

Security begins with our team. Every researcher, engineer, and operator at San Francisco Labs undergoes rigorous background screening prior to employment. All employees participate in mandatory, continuous security awareness training tailored to the unique threats facing AI laboratories, including spear-phishing, social engineering, and operational security (OpSec) best practices.

16. Coordinated Vulnerability Disclosure

We recognize the vital role the independent security research community plays in securing the internet. If you have discovered a potential security vulnerability in our infrastructure, APIs, or models, we strongly encourage you to disclose it to us privately.

Please submit detailed reports, including reproduction steps, via the Contact link provided in the footer. We commit to acknowledging your report promptly, working collaboratively to patch the vulnerability, and refraining from initiating legal action against researchers who conduct their testing in good faith and in accordance with a responsible disclosure framework.