INFORMATION SECURITY
Security is how the platform earns its place.
Bold Penguin moves commercial insurance submissions between agents, brokers, and carriers all day long. This page is the short version of how we protect that data: the architecture, the controls, and how we secure the AI in the platform.
AES
AES-256 encryption at rest across EBS, S3, RDS, and ElastiCache.
SOC 2
Type II audited against theTrust Service Criteria for Security, Availability, Confidentiality, and Privacy
24 / 7
Intrusion detection and centralized logging, evaluated in real time.
SECURITY ARCHITECTURE
Defense in depth, by design.
Bold Penguin architects its products to be nimble and reliable at the same time. Layered controls let the team adapt to new cloud patterns while containing risk and failure at scale.
A culture of security
No control compensates for a workforce that doesn’t take security seriously. So we start with people.
- Background checks before joining, where legally permitted
- Least-privilege provisioning scoped to the role
- Annual security training and ongoing threat communication
- Internal policies documented and published company-wide
- Audited offboarding, confirmed by quarterly review
01
A cloud-first perimeter
Business-critical infrastructure runs entirely in AWS. There is no on-premise hardware to inherit risk from, and no path from the corporate office to production.
- Production runs in AWS; Bold Penguin has no access to physical AWS infrastructure
- The office holds no servers and no site-to-site connection to production
- Network boundaries are enforced in software: security groups, private subnets, no standing inbound access
- Built for a remote-first workforce; the office is not a trust boundary
02
Isolation as the default
Each AWS account is a hard boundary. Blast radius is a design constraint, not an afterthought.
- Multi-account AWS structure with unique, isolated identities
- A change in one account cannot affect resources in another
- Multiple availability zones remove single points of failure
03
Encryption everywhere
Every byte at rest is encrypted with AES-256. Everything in motion rides current TLS. There is no unencrypted tier.
- AES-256 at rest
- TLS 1.2+ during transit
04
DevOps means DevSecOps
Security sits as far left in the SDLC as it will go, because catching an issue earlier usually means never introducing it.
- Infrastructure as code, peer-reviewed like application code
- No change merges without another developer’s approval
- Automated code and dependency scanning in the pipeline
- Production is not hand-edited outside automated deploys
- Federated identity with MFA; RBAC on IAM, SSH, and databases
05
Trust, but verify
Automated tooling paired with manual verification, and an external view that backstops the internal one.
- Annual third-party penetration test
- Internal black-box adversary emulation aligned to MITRE ATT&CK
- Documented incident response and business continuity programs
06
AI & AGENTIC SECURITY
The model reads. Code decides.
Bold Penguin's agentic platform puts language models to work on extraction and analysis — and nowhere near the decisions. Every control below is documented in our developer documentation, not just asserted here.
Deterministic decisions
LLMs extract and analyze. Deterministic code validates, scores, routes, and decides.A model cannot approve its own output or talk its way past a control, because the control is not implemented in the prompt. Requests that violate content guardrails are refused outright.
01
Human in the loop
Review gates fire on explicit tool calls, never on the model’s read of the conversation. Plan confirmation, discrepancy corrections, application consent, carrier selection, and quote selection all pause for a person. The pipeline never auto-proceeds past a gate.
02
Immutable prompts
Orchestrator prompts are version-controlled files, pinned to a release, and never mutated mid-session. Tool-internal prompts stay behind the MCP boundary in their own context, so nothing in a conversation can reach them.
03
Data stays server-side
The conversation carries references; the tools hold the artifacts. Extraction output, application forms, and quote responses are persisted server-side and never pass through model context, so submission data is not sitting in a prompt history.
04
Scoped tool access
No agent sees every tool. Toolsets are declared and filtered per subagent, and MCP servers are process-isolated behind IAM-signed calls. Adjacent servers are unbound by default rather than reachable by default.
05
Provenance on every field
Each run is bound to the user who created it — another user’s credentials get a 403, not a session. Every tool call, validation decision, and answer is logged, and answers are tagged with where they came from: enriched, extracted, or defaulted.
06
BOLD PENGUIN AGNTIC PLATFORM — DESIGN PLATFORM
LLMs extract and analyze. Deterministic code validates, scores, routes, and decides.
ASSURANCE
Audited, regulated, and reviewed.
Independent assurance and the obligations we are measured against. Current reports and certificates are available through the Trust Center.
SOC 2 Type II
Third-party audit of systems and business processes against the Security, Availability, Confidentiality, and Privacy trust services criteria.
AUDITED
NYDFS 23 NYCRR 500
Bold Penguin follows the New York Department of Financial Services cybersecurity requirements for financial organizations.
REGULATED
CCPA / CPRA
Consumer and employee privacy notices maintained, with data subject rights fulfilled inside statutory deadlines.
PRIVACY
TRUST CENTER
The proof lives one click away.
This page is the introduction. Our Trust Center is where the documents are.
TRUSTBOLDPENGUIN.COMSOC 2 Type II report
Penetration test summary
Information Security Policy
Security Architecture whitepaper
Subprocessor list
Business continuity & incident response summaries
Data Processing Addendum
Found something? Tell us.
We extend safe harbor to researchers who report in good faith and stay in scope. Reports go straight to the security team, not to a queue.