<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Darshil Babel</title>
    <description>The latest articles on DEV Community by Darshil Babel (@darshil_babel).</description>
    <link>https://dev.to/darshil_babel</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4244976%2F9f5a340c-6127-492d-ab56-8fffc16286ed.jpg</url>
      <title>DEV Community: Darshil Babel</title>
      <link>https://dev.to/darshil_babel</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/darshil_babel"/>
    <language>en</language>
    <item>
      <title>DPDP Phase 2 has started: a six-month checklist for document-heavy teams</title>
      <dc:creator>Darshil Babel</dc:creator>
      <pubDate>Wed, 18 Nov 2026 05:52:07 +0000</pubDate>
      <link>https://dev.to/darshil_babel/dpdp-phase-2-has-started-a-six-month-checklist-for-document-heavy-teams-5286</link>
      <guid>https://dev.to/darshil_babel/dpdp-phase-2-has-started-a-six-month-checklist-for-document-heavy-teams-5286</guid>
      <description>&lt;p&gt;&lt;strong&gt;The short answer:&lt;/strong&gt; on 13 November 2026 the second phase of India's DPDP rollout began: the consent manager framework (Rule 4) took effect and the Data Protection Board opened registration. Most obligations for Data Fiduciaries apply from &lt;strong&gt;13 May 2027&lt;/strong&gt;. For banks, NBFCs, insurers and other document-heavy teams, the next six months are for finding where personal data sits inside documents, reducing it, and being able to prove what you did.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;This is a practical summary, not legal advice. Check the DPDP Act, 2023 and the DPDP Rules, 2025 for the exact text.&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What changed on 13 November 2026
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Consent managers can now register.&lt;/strong&gt; Indian companies with a net worth of at least Rs 2 crore can apply to the Data Protection Board to operate as consent managers, platforms that let people give, manage and withdraw consent across services. (&lt;a href="https://www.india-briefing.com/news/india-dpdp-consent-manager-registration-47884.html" rel="noopener noreferrer"&gt;India Briefing&lt;/a&gt;, &lt;a href="https://www.tribuneindia.com/news/consent-managers/govt-opens-registration-for-consent-managers-under-dpdp-act-rs-2-crore-net-worth-required" rel="noopener noreferrer"&gt;The Tribune&lt;/a&gt;)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The rest of the clock keeps running.&lt;/strong&gt; The Rules were notified on 13 November 2025 with a phased timeline. The core duties (notices, security safeguards, breach intimation, erasure) apply from 13 May 2027.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Why documents are the hard part
&lt;/h2&gt;

&lt;p&gt;Databases are easy to inventory. Documents are not. KYC copies, salary slips, bank statements, Form 16s, claim forms and scanned applications sit in shared drives, email attachments, loan-origination systems and, increasingly, LLM and OCR pipelines. Many are scans, many are in Hindi, Marathi or other Indian languages, and personal data appears in free text, tables, stamps and handwriting.&lt;/p&gt;

&lt;h2&gt;
  
  
  A six-month checklist for document-heavy teams
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;1. Map personal data inside documents.&lt;/strong&gt; For each document type you collect, list the personal data fields it carries and every system it flows into, including outside processors and AI vendors.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;2. Collect less, and mask at intake.&lt;/strong&gt; Store only the fields the purpose needs. Existing UIDAI requirements already restrict storing full Aadhaar numbers; apply the same thinking to PAN, account numbers and dates of birth on every stored copy.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;3. Build security safeguards into the document flow.&lt;/strong&gt; The Rules list reasonable security safeguards such as encryption, obfuscation or masking, and the use of virtual tokens mapped to personal data, along with access controls and logs (Rule 6). Redacting or masking documents before they are shared or stored is the document-level version of this.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;4. Redact before sharing with processors and AI tools.&lt;/strong&gt; When documents go to a vendor, an OCR service or an LLM, send only what the task needs and remove the rest from both the text and the page image, inside your own environment.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;5. Be ready for breach intimation.&lt;/strong&gt; The Rules require informing the Board and affected people about a personal data breach, with a detailed report to the Board within 72 hours (Rule 7). That is only possible if you know which documents held which personal data.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;6. Keep an audit trail.&lt;/strong&gt; Record what was detected and removed in each document, under which policy and who approved it, so you can show your safeguards rather than describe them.&lt;/p&gt;

&lt;p&gt;Redaction is one safeguard, not the whole of compliance: it does not replace notice, consent, retention and access-control decisions.&lt;/p&gt;

&lt;h2&gt;
  
  
  Test with documents that look like yours
&lt;/h2&gt;

&lt;p&gt;Masking that works on a clean English PDF can miss a value on a skewed phone photo of a Marathi form. GritWorks' free sample packs include fictional Indian-language documents (Hindi, Marathi and Bengali), each clean and degraded, with the expected value and position of every field. &lt;a href="https://gritworks.ai/resources/free-datasets?utm_source=devto&amp;amp;utm_medium=referral&amp;amp;utm_campaign=crosspost-signal-dpdp-phase-2" rel="noopener noreferrer"&gt;Get a free sample pack&lt;/a&gt;.&lt;/p&gt;

</description>
      <category>privacy</category>
      <category>security</category>
      <category>india</category>
      <category>ai</category>
    </item>
    <item>
      <title>Every document you send to an AI vendor is data they now hold: 6 controls</title>
      <dc:creator>Darshil Babel</dc:creator>
      <pubDate>Wed, 18 Nov 2026 05:51:35 +0000</pubDate>
      <link>https://dev.to/darshil_babel/every-document-you-send-to-an-ai-vendor-is-data-they-now-hold-6-controls-55ni</link>
      <guid>https://dev.to/darshil_babel/every-document-you-send-to-an-ai-vendor-is-data-they-now-hold-6-controls-55ni</guid>
      <description>&lt;p&gt;&lt;strong&gt;The short answer:&lt;/strong&gt; every document you send to an AI provider becomes data that provider stores, and a breach at that provider becomes your breach. Before documents leave your environment, send only what the task needs, remove identifiers from both the text and the page image, do that step inside your own perimeter, and treat vendor credentials like production secrets.&lt;/p&gt;

&lt;h2&gt;
  
  
  What happened
&lt;/h2&gt;

&lt;p&gt;Uber is notifying drivers that personal information, including driver's license data, was taken in an October breach at &lt;strong&gt;Fireworks AI&lt;/strong&gt;, a third-party AI vendor. According to state breach filings reviewed by reporters, thousands of drivers are affected (2,567 in Massachusetts, 3,430 in South Carolina and 1,873 in Rhode Island, with other states pending). Attackers used compromised Fireworks credentials to reach data Uber had stored with the vendor; Uber says its own systems were not breached and it has found no misuse so far. Notion separately said the same provider was behind an exposure of AI chat transcripts affecting a small share of its workspaces. (&lt;a href="https://www.yahoo.com/news/us/articles/hackers-stole-uber-drivers-license-185904205.html" rel="noopener noreferrer"&gt;Yahoo News / TechCrunch&lt;/a&gt;, &lt;a href="https://www.ibtimes.com/uber-driver-data-breach-fireworks-ai-3809743" rel="noopener noreferrer"&gt;IBTimes&lt;/a&gt;)&lt;/p&gt;

&lt;p&gt;Public reporting has not explained why the driver records were reachable through the vendor account, so this post is not about what any one company should have done. It is about a pattern every team building on LLMs now shares.&lt;/p&gt;

&lt;h2&gt;
  
  
  The pattern: AI vendors hold more of your data than you think
&lt;/h2&gt;

&lt;p&gt;When you call a model provider, documents and prompts can end up in request logs, batch-job files, evaluation or fine-tuning datasets, caches and support tickets. Each of those copies sits behind the vendor's credentials and the vendor's security, not yours. The more providers you use (model hosts, OCR APIs, agent platforms, observability tools), the larger that surface gets.&lt;/p&gt;

&lt;h2&gt;
  
  
  Six controls for documents that go to AI vendors
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;1. Know what you send.&lt;/strong&gt; List every pipeline that sends documents or extracted text to an outside model, and which document types flow through it: ID documents, pay stubs, bank statements, claims, medical records.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;2. Send only what the task needs.&lt;/strong&gt; If the model is classifying a document or extracting an address, it does not need the license number, date of birth or photo. Drop or mask fields the task does not use.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;3. Redact the image, not just the text.&lt;/strong&gt; ID documents and scans are images. Removing a value from the OCR text while the page image still shows it does not protect it. Remove the pixels, the text layer and the metadata, and check the output by trying to copy, search and extract the redacted areas.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;4. Do the redaction inside your perimeter.&lt;/strong&gt; If you send originals to a cloud redaction service so that you can safely send them to a model provider, you have added another vendor that holds the originals. Run detection and redaction on-premises or in your own cloud account, and send only redacted output.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;5. Use tokens when you need to re-link results.&lt;/strong&gt; Replace identifiers with consistent placeholders (for example &lt;code&gt;&amp;lt;DL_NUMBER_1&amp;gt;&lt;/code&gt;) and keep the mapping in your own systems, so model output can be joined back to the record without the vendor ever seeing the value.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;6. Treat vendor access like production access.&lt;/strong&gt; Scoped keys, rotation, short retention or zero-retention settings where the vendor offers them, and deletion of batch files and datasets you no longer need. This incident began with compromised credentials.&lt;/p&gt;

&lt;p&gt;Be clear about the limits: redaction reduces what a vendor breach can expose, but it is not an access-control decision and it does not replace the vendor's own security. On its own it is also not formal de-identification under rules such as HIPAA.&lt;/p&gt;

&lt;h2&gt;
  
  
  Test it on documents that look like yours
&lt;/h2&gt;

&lt;p&gt;Redaction that works on a clean PDF can miss a value on a skewed phone photo of a license. Test on documents where you know every sensitive value in advance, in the capture conditions you actually receive. GritWorks' free &lt;a href="https://gritworks.ai/resources/free-datasets?utm_source=devto&amp;amp;utm_medium=referral&amp;amp;utm_campaign=crosspost-signal-ai-vendor-breach" rel="noopener noreferrer"&gt;US KYC sample pack&lt;/a&gt; has fictional driver's license and proof-of-address specimens, each clean and degraded, with the expected value and position of every field.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>security</category>
      <category>privacy</category>
      <category>llm</category>
    </item>
  </channel>
</rss>
