Pharma & Life Sciences
Use AI in Pharma and Life Sciences Without Sending Patient Data to the Model
- 01
Identify sensitive values
- 02
Replace with stable tokens
- 03
On API traffic, restore 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 personal data meets AI in a life sciences company
A pharmacovigilance associate has a case narrative due and pastes the adverse event report into a chatbot for a summary. The report has the reporter's name and phone number, the patient's initials and date of birth, a hospital, and a treating physician. A medical writer does something similar with a site monitoring report, and a medical information team drafts a reply to a doctor's letter that names a patient.
Clinical trial data is usually key-coded, so it's tempting to treat it as anonymous. Under GDPR, key-coded data generally stays personal data for anyone who could reach the key, and free-text fields in safety reports and monitoring notes often carry the very details the coding was meant to remove. Health and genetic data are also special-category data under Article 9.
Maskflare applies one rule set across these routes: the desktop agent and browser protection for staff on AI chat, the gateway for the assistants you build, and Exact Data Match or custom rules for subject, site and batch codes in your own formats. It doesn't touch your validated systems, and the masked text keeps the clinical meaning a model needs.
Where pharma & life sciences is actually exposed
5 failure modes specific to this environment, not a restatement of general security advice.
01
Adverse event narratives pasted for summary
A safety case carries reporter details, dates, a hospital and a physician alongside the clinical story. The desktop agent masks those identifiers before the prompt reaches ChatGPT, Claude, Gemini, DeepSeek, Perplexity, Mistral Le Chat and Microsoft 365 Copilot. The answer isn't rewritten, so it may show tokens where it refers to them, and the associate fills them in or keeps them.
02
Key-coded subject IDs with house formats
Subject, randomization and site codes follow your protocol's format, and there's no built-in detector for them. A custom regex rule or Exact Data Match on your subject list masks them, so a code can't be tied back through a provider's logs.
03
Investigator and site correspondence
Monitoring reports and site emails name investigators, coordinators and institutions. The person names pack and contact detectors mask individuals; decide separately whether a site's name is confidential to you, and add it to a dictionary if so.
04
Unpublished trial results and formulas
Masking looks for identifiers. An unblinded efficacy table or a process parameter isn't an identifier and won't be caught without a dictionary or custom rule, so this kind of confidential data needs a policy decision on which tools may receive it.
05
Coding agents in data and bioinformatics teams
Statisticians and engineers point coding agents at real extracts and config files. Hooks for Claude Code, OpenAI Codex, Cursor, GitHub Copilot CLI and VS Code agent mode mask tool input and output, and block prompts containing secrets.
Who this page is for
- Head of Pharmacovigilance
- Wants case processing support from AI without each narrative becoming a personal data transfer.
- Clinical operations lead
- Needs monitoring and site documents summarized without investigator and subject details leaving the company.
- Data protection officer
- Has to justify any transfer of Article 9 health and genetic data, and show what each AI provider receives.
- Quality and computerized systems validation
- Needs to know what sits on the path of GxP-relevant work, and to keep AI use outside validated records unless it has been assessed.
- CISO
- Needs one control for staff AI chat, internal assistants and coding agents, with evidence for audits.
How a team would actually run this
Summarizing a safety case with AI
- 01 Deploy the desktop agent to safety and medical teams with Intune, Jamf or Group Policy. Only AI hosts are routed through Maskflare.
- 02 Apply the GDPR profile or the Swiss nDSG health data profile, plus the country packs for your markets.
- 03 Add a custom rule for your case number format, and upload known subject codes to Exact Data Match. Only keyed hashes leave your environment.
- 04 The associate pastes the narrative and asks for a summary. The provider sees tokens. The summary may show those tokens where it refers to masked details, and the associate fills them in or keeps them, then checks it as they would any draft.
Building a medical information assistant
- 01 Point the assistant's OpenAI or Anthropic SDK at the Maskflare gateway, which is a base-URL change.
- 02 Run new rules in alert mode for two weeks to see what turns up in real enquiries.
- 03 Move rules to mask with maker-checker approval, so a second person signs off each change.
- 04 Keep embeddings on deterministic tokens, so a retrieval index over past enquiries holds tokens rather than names.
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 patient and investigator names, national IDs, dates and health identifiers before a prompt reaches an AI provider, and on API traffic restores them in the answer.
Detection engine
Maskflare Exact Data Match
Masks your own subject, site and case codes from keyed hashes of your records, with no clear values sent.
For your people
Maskflare Workforce
The desktop agent covers safety, clinical and medical teams on AI chat sites, deployed through device management.
For your people
Maskflare User Coaching
A warn rule asks people to confirm or justify before a category such as health data leaves, which suits work where masking alone isn't enough.
Detection engine
Maskflare File Security
Scans study documents and reports, masks plain-text files in place, and strips metadata from images.
For your agents
Maskflare Coding Agents
Keeps extracts, secrets and connection strings out of the prompts and tool calls of coding agents.
The data, and how it's found
Typical data in this sector, the detectors that cover it, and where you'll need your own rules.
| Data | How Maskflare finds it | Profile or rule |
|---|---|---|
| Patient and investigator names | Person names pack and local named-entity layer | GDPR, Swiss nDSG health data, HIPAA Safe Harbor |
| Medical record, encounter and lab numbers | Health pack detectors, using context keywords | HIPAA Safe Harbor, Swiss nDSG health data |
| Clinical and birth dates | Clinical date and date of birth detectors | HIPAA Safe Harbor |
| National and insurance IDs, such as Swiss AHV | 47 country packs with checksum and format validation | GDPR, Swiss nDSG health data |
| Diagnoses and genetic data in free text | Special categories pack: masked on the default starter, alert on Monitor first | Your policy decision |
| Trial subject, site and batch codes | No built-in detector: custom regex or Exact Data Match | Your own rules |
| Medical device identifiers | UDI and device serial detectors | HIPAA Safe Harbor |
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 and key-coded data
- Health and genetic data are special-category data, and key-coded data generally stays personal data for anyone who can reach the key. Reversible tokens are pseudonymized data for you too, so your legal basis still applies.
- Clinical Trials Regulation (EU) 536/2014 and ICH E6
- Set expectations for protecting subject data and confidentiality in trials. Masking is a data-minimization control you can document; it doesn't replace your sponsor or site duties.
- EU GMP Annex 11 and draft Annex 22
- Annex 11 covers computerised systems in GMP. Annex 22 on AI was published as a draft for consultation in 2025, and we couldn't confirm a final version as of November 2026. Maskflare isn't a validated GxP system, so keep it out of your validated record path unless you've assessed it.
- FDA 21 CFR Part 11 and HIPAA
- Part 11 concerns electronic records and signatures, which Maskflare doesn't manage. Where HIPAA applies, the HIPAA Safe Harbor profile covers the identifiers it names; expert determination and your BAA position stay with you.
- Swiss Human Research Act and revised FADP
- Swiss research rules and data protection law apply to health-related research data. The Swiss nDSG health data profile turns on the relevant detectors. Your ethics and legal review stays with you.
What Maskflare does not do
- Masking removes identifiers, not facts. A rare disease at a named small site can still point to a person, so judge what context you send.
- No built-in detector exists for subject, site or batch codes, because they differ by protocol. You add them as custom rules or Exact Data Match.
- It isn't a validated GxP system, an eTMF, or a safety database, and it doesn't validate what the AI writes.
- Documents, including scanned pages read with OCR, are scanned under your document policy and recorded or refused, not rewritten. In browser apps, picture uploads pass unread or are refused, and uploads that go to a separate storage host are refused.
- SOC 2 Type II and ISO 27001 audits are in progress.
5 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 subject code, a name or a date tied to a clinical story, or only tokens?
- 02Can it match our own subject and site codes without holding them in clear text?
- 03What happens to confidential data that isn't personal, such as unblinded results?
- 04Where does detection run, and does any third-party AI service read the text?
- 05Does the tool sit on the path of any GxP-relevant record, and how do we assess that?
Workflow guides for this industry
How a team applies each of these, including the contrast with the tool it replaces.
Pharma & Life Sciences questions we get asked
Is key-coded trial data anonymous?
Generally not under GDPR, for anyone who can reach the key. Treat it as personal data unless your legal team has concluded otherwise, and check free-text fields, which often contain details the coding was meant to remove. This is not legal advice.
Can Maskflare detect our subject and site codes?
Not out of the box, since each protocol has its own format. Add a regex rule for the format, or upload known codes to Exact Data Match, where only keyed hashes leave your environment.
Can safety teams paste adverse event reports into ChatGPT?
With the desktop agent installed, names, dates, IDs and contact details are masked before the report reaches ChatGPT. The answer may show tokens where it refers to them, and the person fills those in or keeps them. The clinical description is still sent, so decide with your DPO and safety lead which case types are in scope.
Does Maskflare protect unpublished results or formulas?
Only if you add terms or patterns for them, because masking looks for identifiers. For this kind of confidential data, a block rule or a decision on approved tools is the better control.
Is Maskflare validated for GxP use?
No. It controls what data reaches AI providers, and it doesn't create or manage GxP records. Assess it under your own procedures if it would sit on the path of GxP-relevant work.
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.
See it on your own data
Start a free trial or talk to us with the identifiers and AI workflows your pharma & life sciences 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 →