Insurance
Use AI on Claims and Underwriting Without Sending Policyholder Data to the Model
- 01
Identify sensitive values
- 02
Replace with stable tokens
- 03
Restore only inside your boundary
574
Built-in detectors for personal data and secrets
47
Country packs, checksum-validated where the format has one
9
AI providers behind one gateway, plus the desktop agent and proxy
Where policyholder data meets AI at an insurer
A claims handler at 4 p.m. has a motor claim with a police report, two repair quotes, a physiotherapy report, and fourteen emails. Pasting the lot into a chatbot and asking for a timeline is the obvious move. That paste carries the claimant's name, date of birth, policy number, IBAN for the payout, and a description of a neck injury.
Underwriters do the same with broker submissions and medical questionnaires. Customer service drafts replies to complaints. Actuarial and data teams build retrieval over policy wordings and claims history. Health details turn up in all of it, and health data is special-category data under GDPR Article 9.
Maskflare applies one rule set to all of these: the desktop agent and browser protection for staff on AI chat, the gateway for the tools you build, and Exact Data Match for the policy and claim numbers only your systems know.
Where insurance is actually exposed
5 failure modes specific to this environment, not a restatement of general security advice.
01
Claims files pasted for summarization
A claim summary request carries names, addresses, policy numbers, payout IBANs, and medical findings. The desktop agent masks them before the prompt reaches ChatGPT, Claude, Gemini, DeepSeek, or Perplexity, and puts the real values back where the summary repeats them.
02
Medical reports inside claims
Injury, disability, and health claims include doctors' reports with diagnoses and health identifiers. The healthcare pack and person names mask the identifiers; the clinical description itself stays, so the model can still assess it without knowing whose it is.
03
Policy and claim numbers with house formats
Every insurer formats policy and claim numbers differently. Upload them as keyed hashes with Exact Data Match, or add a regex rule, so they are masked as reliably as an IBAN.
04
Broker submissions and attachments
Underwriting submissions arrive as PDFs, spreadsheets, and forwarded emails. Document scanning reads their text and blocks or records them per policy; plain-text and CSV files are masked in place.
05
Retrieval over policy wordings and claims history
Embedding past claims for retrieval puts personal data into a vector store. Embeddings requests through the gateway use deterministic tokens and fail closed, so the index holds tokens instead of names.
Who this page is for
- Head of Claims
- Wants handlers to get AI help on long files without each paste becoming a data incident.
- Chief Underwriting Officer
- Needs underwriters to use AI on submissions and medical questionnaires with health data kept out of the provider's logs.
- Data protection officer
- Has to justify any transfer of Article 9 health data and show what each AI provider receives.
- CISO
- Needs one control across staff AI chat, internal tools, and the claims platform's own AI features.
- Outsourcing and third-party risk
- Has to document each AI provider as an outsourcing or ICT third-party arrangement, including the data it gets.
How a team would actually run this
Summarizing a claims file with AI
- 01 Deploy the desktop agent to claims handlers with Intune, Jamf, or Group Policy. Only AI hosts are routed through Maskflare.
- 02 Turn on the GDPR profile, the healthcare pack, and the country packs for your markets, so IBANs and national IDs are checksum-validated.
- 03 Upload policy and claim numbers to Exact Data Match from your policy admin system with `maskflare edm upload`; only keyed hashes leave your environment.
- 04 The handler pastes the file and asks for a timeline. The provider sees tokens; the handler sees the timeline with real names and numbers restored.
Building an underwriting assistant
- 01 Point the assistant's OpenAI or Anthropic SDK at the Maskflare gateway: a base-URL change.
- 02 Start new rules in alert mode for two weeks to measure what turns up in real submissions before you mask or block.
- 03 Move rules to mask with maker-checker approval, so a second person signs off on each change.
- 04 Export the compliance coverage report as CSV or PDF for the outsourcing file.
Which modules apply, and why
Each links to the module page, where the boundaries and development status are set out in full. Or start at the Maskflare platform overview.
Detection engine
Maskflare Detection Engine
Detects and masks policyholder names, national IDs, IBANs, health identifiers, and dates of birth before a prompt reaches an AI provider, and restores them in the answer.
Detection engine
Maskflare Exact Data Match
Masks your own policy, claim, and customer numbers from keyed hashes of your records, with up to 10 datasets and no clear values sent.
For your people
Maskflare Workforce
The desktop agent covers claims handlers and underwriters on AI chat sites, deployed through device management and fail-closed by default.
For your people
Maskflare Browser Protection
Masks text typed into AI chat sites and text uploads such as CSV exports, and restores the values where the answer repeats them.
Detection engine
Maskflare File Security
Scans claim documents and broker submissions in a sandbox, masks plain-text files in place, and strips location metadata from claim photos.
For your apps
Maskflare AI Gateway
Puts underwriting assistants and claims retrieval behind the same rules, with deterministic embeddings that fail closed.
Regulatory context
Readiness and relevance, not certification. Nothing here is a claim to hold an audit or authorization we do not have.
- GDPR Art. 9
- Health data in claims and underwriting is special-category data. Masking limits what reaches an AI provider; reversible tokens remain pseudonymized personal data for you, so your legal basis still applies.
- EIOPA AI Opinion (EU)
- EIOPA published its Opinion on AI governance and risk management on 6 August 2025, applying existing Solvency II and IDD requirements to AI. Masking is one data-management control you can document under it.
- FINMA Guidance 08/2024 (Switzerland)
- FINMA's December 2024 guidance on AI governance covers data quality, outsourcing, and risk management at supervised insurers. Maskflare can support the outsourcing and data parts; it doesn't produce an AI inventory or model validation.
- Solvency II and DORA
- AI providers, and Maskflare, are outsourcing or ICT third-party arrangements. Masking changes what you record about each provider's data access.
- IDD and EU AI Act
- The IDD sets conduct duties for distribution, and the AI Act treats AI for risk assessment and pricing in life and health insurance as high-risk. Maskflare doesn't assess or classify AI systems.
What Maskflare does not do
- Masking removes identifiers, not facts. A rare injury in a small town can still point to a person, so judge what context you send.
- Scanned PDFs aren't read, and text in images is read only when image scanning is on (classic OCR, best effort, API requests only).
- No fraud detection, claims scoring, or pricing models. Maskflare controls what data reaches AI, not what the AI decides.
- SOC 2 Type II and ISO 27001 audits in progress.
4 questions to ask any vendor here
Including us. If our answer is worse than someone else's, you should know that before you buy, not after.
- 01Does the provider ever receive a policyholder name, policy number, or diagnosis tied to a name, or only tokens?
- 02Can it match our own policy and claim numbers without holding them in clear text?
- 03How are attached claim documents and photos handled, and which file types does it read?
- 04Where is it hosted, what does it store, and what certifications does it hold today?
Workflow guides for this industry
How a team applies each of these, including the contrast with the tool it replaces.
Insurance questions we get asked
Can claims handlers paste medical reports into ChatGPT?
With the desktop agent installed, names, dates of birth, national IDs, and health identifiers in the report are masked before it reaches ChatGPT, and restored where the answer repeats them. The medical description itself is sent, so decide with your DPO which claim types are in scope.
How do you detect our policy numbers?
Either as a regex rule for your format, or with Exact Data Match: you upload keyed hashes of your policy and claim numbers with the Maskflare CLI, and only those exact values are masked. The clear values never leave your environment.
Does this cover claims photos?
Image metadata such as EXIF GPS coordinates is stripped. Images can be allowed, blocked, or scanned per policy; text inside images is read only when image scanning is on, on a best-effort basis.
Where is Maskflare hosted?
In the EU by default, with on-premises deployment available for enterprise. Prompts and responses are never stored; masked values are kept encrypted for 30 days so answers can be restored.
Does masking make our AI use compliant with the EIOPA Opinion or FINMA guidance?
No single tool does. Masking is a data-minimization control you can document; governance, inventories, testing, and accountability stay with you. This is not legal advice; check current guidance as of October 2026.
See it on your own data
Start a free trial or talk to us with the identifiers and AI workflows your insurance team actually handles. Bring the case that makes it hard, because that is the part worth testing.
See it on your own data
Try Maskflare free.
Bring your hardest prompts.
28-day free trial or 10,000 requests, whichever comes first. Point one app at the gateway or enroll a few laptops,
and watch what gets masked. Or talk to us first.
- SOC 2 Type II: In progress
- ISO 27001: In progress
- EU hosting by default · on-premises available
- Prompts and files never stored
- Never used to train AI models
- GDPR · nDSG · DPA available
- Trust center →