Skip to main content
Back to Blog
SecurityInsights6 min read-

Ur AI Achieves SOC 2 Type 2 and GDPR Compliance: What Those Words Actually Prove

Ur AI has completed its SOC 2 Type 2 examination and its GDPR compliance program, joining the ISO 27001:2022 certification we received in January. What each framework proves, what none of them prove, and how to check any AI vendor for yourself.

Sandeep Yella

Sandeep Yella

Founder & CEO

Ur AI Achieves SOC 2 Type 2 and GDPR Compliance: What Those Words Actually Prove

As of September 3, 2026, Ur AI has completed its SOC 2 Type 2 examination and its GDPR compliance program. Together with the ISO 27001:2022 certification we received in January 2026, enterprise buyers can now verify three separate things about us: that we operate a certified information security management system, that an independent examination covered a multi-month window rather than a single day, and that our handling of personal data is governed by published policies. The SOC 2 Type 2 report and the ISO 27001:2022 certificate are both available on request at trust.ur-ai.net, next to the 75 controls we monitor continuously and the full list of every subprocessor that touches customer data.

Readiness work began in February. The observation window opened on March 27, 2026, and from that day every access review, every change approval, every offboarding checklist, and every quarterly test became evidence that either existed or did not. Evidence cannot be backfilled, and six months is a long time to be graded on consistency.

What SOC 2 Type 2 actually proves (and what it does not)

A SOC 2 report is an attestation, not a certificate. It is an examination performed by a licensed CPA firm against the AICPA's Trust Services Criteria, which cover five categories: security, availability, processing integrity, confidentiality, and privacy. Nobody hands out a SOC 2 badge. What exists is a report.

The difference between the two report types is time. A Type 1 asks whether controls were suitably designed on a given date. A Type 2 asks whether those same controls operated effectively across a defined observation window, typically six to twelve months. A control that is designed well in March and quietly stops running in June does not survive a Type 2. It appears in the report as an exception, in writing, for every customer to read.

A Type 1 asks whether you built the thing. A Type 2 asks whether you lived with it.

The company being examined chooses its own scope. Vendors rarely volunteer that, and it means two SOC 2 reports can describe very different amounts of work. When you receive one, four things determine what it is worth: which trust services categories were in scope, how long the observation window ran, whether the auditor noted any exceptions, and which subservice organizations were carved out. A logo on a website tells you none of that. Ask for the report.

Three frameworks, three different claims

FrameworkWhat it isWho issues itWhat it demonstrates
ISO/IEC 27001:2022Certification of an information security management systemAn accredited certification body (ours was Bureau Veritas Japan)That a documented management system for security exists and is audited on a recurring cycle
SOC 2 Type 1Attestation reportA licensed CPA firmThat controls were suitably designed as of one specific date
SOC 2 Type 2Attestation reportA licensed CPA firmThat controls operated effectively throughout a defined observation window
GDPRA regulation, not a credentialNo one. Compliance is demonstrated, not awardedThat documented obligations for personal data are met and can be evidenced on request

The three are complementary rather than interchangeable. ISO 27001 describes the management system, SOC 2 Type 2 evidences its operation over time, and GDPR governs personal data specifically.

What GDPR compliance means when your software reads documents

There is no such thing as a GDPR certificate. GDPR is European law, and compliance is demonstrated through documented obligations rather than awarded by a body. So when a vendor says they are GDPR compliant, ask to see the policies, the records, and the subprocessor list.

Three articles carry most of the practical weight. Article 32 requires appropriate technical and organisational measures, and names encryption, ongoing confidentiality and resilience, restoration of access after an incident, and regular testing of whether those measures actually work. Article 28 governs processors: a processor may act only on documented instructions, must have authorisation before engaging a sub-processor, must impose the same obligations on that sub-processor, and remains fully liable if the sub-processor fails. Article 83 sets the ceiling for getting this wrong at 20 million euros or four percent of total worldwide annual turnover, whichever is higher.

The European Commission has adopted an adequacy decision for Japan, so personal data can flow from the EEA to Japan without additional transfer safeguards. For a Tokyo company with a documented GDPR program, that removes a step European counterparties usually expect to negotiate.

A document intelligence engine sees whatever is inside the file. A construction submittal carries names and professional stamps. A logistics manifest carries driver details. A retail supply agreement carries signatories. An HR folder is personal data end to end. You do not get to decide in advance that a workload contains no personal data, which is why the Article 32 measures have to be properties of the product: encryption at rest for the datastores, encrypted transmission over public networks, privileged access restricted to a business need, and deletion that actually happens when a customer leaves.

You cannot promise a document parser will never see personal data. You can only build it as though it always will.

What the program actually took

We run continuous control monitoring through Vanta, which is why our trust center shows a live status rather than a PDF from last year. Seventy-five controls are monitored, grouped into five areas.

75 monitored controls, by area

Control areaControlsExamples
Internal security procedures35Change management, SDLC, configuration management, continuity testing, board oversight
Infrastructure security21Unique production authentication, restricted database and firewall access, revocation on termination
Organizational security11Background checks, confidentiality agreements, security awareness training, password policy
Product security5Encryption at rest, encrypted transmission, vulnerability monitoring, annual penetration testing
Data and privacy3Data classification, retention and disposal, customer data deleted on exit

Live status for every control is published at trust.ur-ai.net.

Underneath the controls sit the documents. Eight data protection policies were written, approved, and published: Information Security, Access Control, Cryptography, Data Management, GDPR Compliance, GDPR Incident Response, Business Continuity and Disaster Recovery, and our Privacy Policy. Each one has to be more than a file. Access Control means access reviews and termination checklists completed inside an agreed SLA. Business Continuity means the plan is tested annually, not written annually.

  • Security awareness training completed within thirty days of hire and at least annually after that, for every employee
  • Background checks and signed confidentiality agreements for employees, with the code of conduct extended to contractors
  • Every production change authorized, documented, tested, reviewed, and approved before it shipped
  • Access to production systems, databases, firewalls, and deployment restricted to a business need, and revoked inside an SLA at termination
  • Penetration testing at least annually, with remediation tracked against agreed SLAs
  • Business continuity and disaster recovery plans tested, not merely maintained
  • Annual control self-assessments, plus a board briefing on cybersecurity and privacy risk
  • Nine subprocessors reviewed and published, from cloud and version control through to the model providers

Designing a control is the easy part. The hard part is keeping it running for months while the product ships, the team grows, and the infrastructure changes underneath it. There were weeks when the compliance backlog competed directly with the product backlog, and the compliance backlog won.

Why this matters more for AI vendors than for ordinary software

Most enterprise software holds records that someone already classified: a customer row, an order, a support ticket. Document AI is different. It ingests raw files that nobody triaged first, and it passes their contents to models. That changes which questions are worth asking, starting with which subprocessors see the file.

  • Are my documents used to train models, including by any model provider you route to? Get the answer in writing, then check it against the published subprocessor list.
  • Which subprocessors process my data, and where do they operate? A vendor that will not publish that list cannot satisfy Article 28 for a European counterparty.
  • What is deleted when I leave, and how would you prove it? Deletion on exit should be a monitored control, not a support ticket.
  • Can I have the SOC 2 report rather than the badge? The badge is artwork. The report carries the scope, the observation window, and any exceptions.

What you can verify today

Our trust center at trust.ur-ai.net publishes the frameworks we hold, the live status of all 75 controls, the eight data protection policies, and every subprocessor we use. The ISO 27001:2022 certificate and the SOC 2 Type 2 report are both available there on request.

Inside the products, these controls line up with commitments we had already made. Nebula, our document intelligence engine, keeps per-tenant data isolation, authenticates with per-customer API keys, records a per-job audit trail, and does not use customer documents to train models. Specter, which is built on top of Nebula, cites every finding back to its source page so a reviewer can check the reasoning instead of trusting it. Compliance frameworks show that a vendor operates carefully. Traceable outputs let a construction team, a logistics operator, or an investment committee defend a decision months later.

Certifications tell you a vendor is careful with your documents. Auditability tells you whether you can defend the decision you made from them.

A SOC 2 Type 2 report always describes a window that has already closed. The next one opened the day this one ended, and the control status stays live at trust.ur-ai.net in between.

Ur AI is ISO 27001:2022 certified, has completed a SOC 2 Type 2 examination, and operates a documented GDPR compliance program. Request the report, the certificate, or a security review at trust.ur-ai.net.

SOC 2GDPRComplianceEnterprise SecurityTrust Center

Frequently Asked Questions

Is Ur AI SOC 2 compliant?

Yes. Ur AI completed its SOC 2 Type 2 examination in September 2026, following an observation window that opened on March 27, 2026. Ur AI has also held ISO/IEC 27001:2022 certification since January 2026 and operates a documented GDPR compliance program. The SOC 2 Type 2 report and the ISO 27001:2022 certificate are both available on request at trust.ur-ai.net, alongside the live status of 75 monitored controls and the full subprocessor list.

What is the difference between SOC 2 Type 1 and SOC 2 Type 2?

A SOC 2 Type 1 report evaluates whether a service organization's controls were suitably designed as of a single date. A SOC 2 Type 2 report evaluates whether those same controls also operated effectively across a defined observation window, typically six to twelve months. Type 2 is the stronger evidence, because it shows sustained operation rather than a point-in-time snapshot, and any control that lapsed during the window is recorded in the report as an exception.

Is there such a thing as a GDPR certificate?

No. GDPR is European Union law rather than a certification scheme, so no body issues an official GDPR certificate. Organizations demonstrate compliance through evidence: documented policies, a lawful basis for each processing activity, the Article 32 security measures such as encryption and regular testing, Article 28 processor terms with published subprocessors, and a working process for data subject requests and breach notification. When a vendor claims GDPR compliance, ask to see those documents.

What should I ask an AI vendor about data security?

Ask four things. Whether your documents are used to train models, including by any model provider the vendor routes to. Which subprocessors process your data, and where they operate. What is deleted when you stop using the service, and how that deletion is proven. And whether you can receive the actual SOC 2 report rather than a badge, since only the report states the scope, the observation window, and any exceptions the auditor recorded.