Paanini, Inc. dba JIFFYAI
Every control, report and sub‑processor behind the platform — in one place.
JIFFYAI automates document and workflow operations for wealth managers, broker-dealers and banks, including ETL from custodial data sources. This is the evidence your review needs, with the dates attached.
Attestations reviewed Sep 25, 2026 · US data residency · SOC 1 & SOC 2 Type II
Attestations
What we hold, who issued it, and through when.
Independent audits and certifications are listed separately from self-attested compliance positions, because they are not the same thing.
ISO/IEC 27001:2022
CurrentCertification
- Valid through
- Jun 30, 2027
- Last certified
- 2024
Information security management certification
SOC 2 Type II
CurrentIndependent audit
- Report period
- 2025
- Next
- Q4 2026
Annual SOC 2 Type II examination
SOC 1 Type II
CurrentIndependent audit
- Report period
- 2025
- Next
- Q4 2026
Annual SOC 1 Type II examination
HIPAA
CurrentSelf-attested
- Report period
- 2025
- Status
- Annual assessment
Annual HIPAA compliance assessment
GDPR / UK GDPR
CurrentSelf-attested
- Report period
- 2025
- Status
- Annual assessment
Annual GDPR / UK GDPR compliance assessment
ISO/IEC 42001:2023
Audit under wayCertification
- Status
- Audit: August 2026
AI management system certification — audit August 2026
CSA STAR
PlannedCertification
- Status
- Q3/Q4 2026
Cloud Security Alliance STAR — planned Q3/Q4 2026
Not held
Stated plainly
- ISO 27701
- Programme aligned, not certified
- PCI DSS
- Not applicable — no cardholder data
- FedRAMP
- Not pursued
Data and AI
Where your data goes when the platform runs AI.
Three inference paths, three providers, three separate sets of terms — plus what leaves the platform when an agent uses a tool. Each is named here rather than described as one pipeline.
Anthropic Claude
Primary LLM inference
- Via
- AWS Bedrock, VPC private endpoint
- Region
- AWS us-east-1, DR in a second US region
- Guardrails
- Bedrock Guardrails — moderation, topic blocking, PII and ePHI filtering, grounding checks
- Terms
- Bedrock enterprise terms — no training on submitted data
Google Gemini
Document intelligence
- Via
- Google Cloud Vertex AI — not through Bedrock
- Region
- Google Cloud, United States
- Guardrails
- Vertex AI safety configuration, configured separately
- Terms
- Vertex AI enterprise terms with a BAA in place
AssemblyAI
Speech transcription — Advisor Assistant
- Via
- Direct API, outside the Bedrock boundary
- Scope
- Only where the Advisor Assistant is enabled
- Guardrails
- Not applicable — transcription, not generation
- Terms
- Executed DPA
Data leaving the platform
Where an agent uses a search or market-data tool, the query it composes is sent to that provider — Tavily for web search, Finnhub and Alpha Vantage for market and reference data. These are read-only sources: the query goes out, customer records do not. All three are named on the sub-processor list rather than treated as infrastructure.
Agentic controls
See every provider that may process customer data- Tool allowlisting — An agent cannot discover or invoke a capability that is not registered; adding one requires AI Review Board approval.
- Read-only by default — Write and external-action permissions are a separate explicit grant, never inherited.
- Human confirmation — Reads proceed within rate limits; anything with external effect stops for a person.
- Immutable system prompt — A fixed prompt layer the model cannot alter, applied on both inference paths.
- Signed action intent — Every agentic action writes an audit record of who, what, when and why.
- AI Review Board approval — Registering a new tool requires a full AIRB review before an agent can reach it.
For regulated firms
Written for the questions your examiner will ask.
Amended Regulation S-P vendor-oversight obligations have applied to all advisers since 3 June 2026. These are the answers your due-diligence file needs.
SOC 1 Type II
For your auditor, not just your CISO
SSAE 18 AT-C §320 report on controls relevant to user entities' internal control over financial reporting, issued annually by an independent CPA firm.
Tenant isolation
Tenant context is derived server-side
Never taken from a client-supplied parameter — not a host header, not a query string. Five enforcement layers. Any finding that crosses a tenant boundary is treated as Critical regardless of the data exposed.
Data residency
United States only
Stored and processed in the AWS region designated for your tenant, replicated to a second US region for disaster recovery. Other geographies are provisionable and given contractual effect in the DPA.
Immutable backups
S3 Object Lock, Compliance Mode
Backups cannot be deleted or overwritten by any user or automated process — including root accounts — until the retention lock expires. Aurora point-in-time recovery over a rolling 35-day window.
Retention and exit
30 days to deletion, certificate in 10
On termination or written request, data is returned machine-readable or securely destroyed within 30 days, with a written certificate of deletion within 10 business days. Regulated financial and healthcare backups are held 7 years.
Records retention
SEC 17a-4 / FINRA 4511 — scoping under legal review
Whether JIFFYAI acts as a system of record, or processes records that remain in your designated repository, is being determined with counsel. We would rather say that than give you an answer you cannot rely on.
Evidence
36 documents, and you can see the list before you ask.
The privacy policy and high-level statements are open to everyone. Reports and policies are released under a click-through NDA. Penetration test and internal risk reports are granted case by case.
- Certifications & attestations5
- Security policies13
- AI governance4
- Platform & architecture9
- Resilience & continuity2
- Legal & contractual2
- Assessments & overviews1
Getting the full library
- 1Verify your work emailA one-time code. There is no account to create.
- 2Sign the NDA onlineClick-through, in the browser. No PDF round-trip and no sales call.
- 3Download everything at onceA single signed bundle, or individual files.
Every bundle ships with an RSA-signed manifest and SHA-256 hashes, so you can prove the files came from trust.jiffy.ai unaltered.
JIFFYAI employee? Sign in with your Microsoft account from the top navigation for full access — no request and no NDA.
Questionnaire
155 questions, already answered.
16 domains, including one written specifically for financial services: SEC and FINRA handling, FCA and PRA, SR 11-7 and SS1/23 model risk, concentration risk, nth-party risk, data residency and audit rights. Answers cite the policy and evidence document behind them, and export to Excel in one click.
Where an answer is Partial, it says so and tells you who to ask. Better that you find it here than in week three of the review.
Controls
How the platform is actually run.
Encryption
AES-256 at rest under customer-managed keys in FIPS 140-2 validated HSMs. Key material is generated in KMS and never exported or handled by personnel; key administration and key use are separate entitlements. TLS 1.3 for external connections, TLS 1.2 minimum.
Access governance
Single sign-on through Microsoft Entra ID for corporate and platform systems. MFA mandatory for all privileged, cloud infrastructure, email and repository access. Quarterly reviews across AWS, Kubernetes, ArgoCD and Vault, including service accounts and AI agents. Revocation within 24 hours of termination.
Penetration testing
Annual, by an independent firm with no other commercial relationship, under CREST and PTES, with vendor rotation. Most recent: 21 May – 18 June 2026, with no Critical or High severity findings.
Change management
All infrastructure defined as code in Terragrunt and Terraform, deployed via GitOps. There is no manual change path to production. AWS Config monitors continuously for drift.
Resilience
Tier 1 services target RTO under 4 hours and RPO under 1 hour. Quarterly tabletop exercises and a full cross-region failover drill annually — most recently June 2026, next scheduled June 2027.
Detection and response
AWS Security Hub, GuardDuty and CloudTrail with 24/7 on-call escalation, plus self-hosted Langfuse for AI observability. For High and Critical incidents: MTTD under 15 minutes, MTTR under 25 minutes.
Published gap register
What we haven't done yet.
Our own architecture overview puts it this way: a reviewer who discovers a gap that a document has quietly stepped around will reasonably question everything else in it. So here are the open items, with their compensating controls, before you find them.
- AI red teaming — Adversarial testing against the OWASP LLM Top 10 has not been performed. The June 2026 assessment covered the application, API surface, cloud configuration and tenant isolation, but not AI attack paths.Next cycle
- Prompt-injection testing — Targeted adversarial testing not yet performed. Mitigated by the immutable system prompt layer and tool allowlisting.Planned
- Bias and fairness testing — No formal programme yet.Planned
- Data loss prevention — Coverage is partial rather than complete.In progress
- Audit log retention — Application audit logs are retained six months; extension of the security and audit trail to three years is under way.In progress
- Cyber recovery exercise — Distinct from the annual DR failover drill; not yet run.Planned
Transparency
Who else touches the data, and how to reach us.
Sub-processors
Public — no NDA requiredEvery third party that may process customer data, with its category, activity and country — including the AI providers on each inference path. Change history included, so you can see what moved and when.
Security contact
For vulnerability reports, questions about the security programme, and vendor assessments.
ciso@jiffy.ai
Platform status
Live availability and incident history, independently hosted.
status.jiffy.ai
Stay informed
Get notified when JIFFYAI achieves new certifications, publishes updated reports, or adds documents to the Trust Center.