Data Privacy in Cross-Platform Collaboration

This Stanford study examined how six major AI companies (Anthropic, OpenAI, Google, Meta, Microsoft, and Amazon) handle user data from chatbot conversations.  Here are the main privacy concerns. 👀 All six companies use chat data for training by default, though some allow opt-out 👀 Data retention is often indefinite, with personal information stored long-term 👀 Cross-platform data merging occurs at multi-product companies (Google, Meta, Microsoft, Amazon) 👀 Children's data is handled inconsistently, with most companies not adequately protecting minors 👀 Limited transparency in privacy policies, which are complex and hard to understand and often lack crucial details about actual practices Practical Takeaways for Acceptable Use Policy and Training for nonprofits in using generative AI: ✅ Assume anything you share will be used for training - sensitive information, uploaded files, health details, biometric data, etc. ✅ Opt out when possible - proactively disable data collection for training (Meta is the one where you cannot) ✅ Information cascades through ecosystems - your inputs can lead to inferences that affect ads, recommendations, and potentially insurance or other third parties ✅ Special concern for children's data - age verification and consent protections are inconsistent Some questions to consider in acceptable use policies and to incorporate in any training. ❓ What types of sensitive information might your nonprofit staff  share with generative AI?  ❓ Does your nonprofit currently specifically identify what is considered “sensitive information” (beyond PID) and should not be shared with GenerativeAI ? Is this incorporated into training? ❓ Are you working with children, people with health conditions, or others whose data could be particularly harmful if leaked or misused? ❓ What would be the consequences if sensitive information or strategic organizational data ended up being used to train AI models? How might this affect trust, compliance, or your mission? How is this communicated in training and policy? Across the board, the Stanford research points that developers’ privacy policies lack essential information about their practices. They recommend policymakers and developers address data privacy challenges posed by LLM-powered chatbots through comprehensive federal privacy regulation, affirmative opt-in for model training, and filtering personal information from chat inputs by default. “We need to promote innovation in privacy-preserving AI, so that user privacy isn’t an afterthought." How are you advocating for privacy-preserving AI? How are you educating your staff to navigate this challenge? https://lnkd.in/g3RmbEwD

Incorporating Data Privacy Clauses in NDAs 🔐 As someone deeply involved in data protection, I have seen firsthand how critical it is to protect sensitive information in our collaborations. In today’s landscape, integrating robust data privacy clauses into Non-Disclosure Agreements (NDAs) is no longer optional—it's essential. Why This Matters: 1. Regulatory Compliance: With regulations like GDPR and CCPA shaping our practices, we must ensure our NDAs reflect these legal requirements. I've witnessed the repercussions of non-compliance, and it's not something any organization can afford. 2. Data Classification: Clearly defining what sensitive data looks like is crucial. For example, specifying categories like PII or financial data helps everyone understand what’s at stake. 3. Access Controls: Establishing who can access sensitive information—and under what conditions—helps uphold the principle of least privilege. I’ve found that clarity here builds trust among all parties involved. 4. Breach Notification: It’s vital to have a breach notification protocol outlined in the NDA. Knowing how to respond swiftly can make all the difference in minimizing damage. 5. Data Transfer: In our globalized world, addressing cross-border data transfers in NDAs ensures we remain compliant with international standards. By embedding these technical aspects into our NDAs, we reinforce our commitment to data integrity and privacy. It’s not just about legal compliance; it’s about cultivating trust in every partnership. Let’s prioritize data privacy in our agreements and foster a culture of accountability in our industry. #DataPrivacy #NDA #LegalCompliance #DataSecurity #RiskManagement #cybersecurity #dataprotection

I’ve been spending the start of the New Year deeply focused on studying for the Certified Data Privacy Solutions Engineer (CDPSE) exam. With my growing work in AI, automation, and platform integrations, I’ve found myself questioning everything, especially how modern AI-driven platforms and third-party integrations impact an organization’s security posture and privacy risk. One topic that comes up again and again: API keys. What happens when a platform has weak or poorly maintained code? ~ Platform vulnerabilities can be exploited • Attackers may bypass controls, access secrets in memory, or manipulate workflows. ~ Your API keys and data can still be exposed • Even if you followed best practices, compromised platform code can leak keys, tokens, or sensitive data being processed. ~ The blast radius depends on your controls • Scoped keys, least privilege, IP restrictions, and short-lived tokens reduce impact. Broad, long-lived keys dramatically increase risk. ~ Shared responsibility still applies • The provider owns platform security. You own configuration, access management, and data protection. Regulators evaluate your safeguards. ~ Trust and compliance exposure is real • Incident response, customer notifications, contract impacts, and regulatory scrutiny can follow, even when the bug wasn’t yours. What strong security-minded organizations do differently: ~Use least-privileged, short-lived API keys ~Rotate keys regularly ~Monitor logs for abnormal usage ~Isolate sensitive processing ~Maintain vendor risk management and exit strategies Bottom line: When a platform fails, your controls determine whether it’s a contained incident, or a major breach. This is where privacy engineering, AI governance, and security architecture intersect. And it’s exactly why understanding how things are built matters just as much as what they do. CISM | CDPSE | AAISM Just a girl working on becoming a triple threat… Security. Privacy. AI. 💅🏽🔐🤖 Happy New Year!! 🎉

I see this happening too often. Advertisers, publishers, agencies, and data providers—rushing into Data Clean Room (#DCR) decisions with minimal due diligence. Not all DCRs are the same. Their #privacy frameworks, #interoperability, #security mechanisms, and #governance models differ significantly. Yet, when the time comes to choose one, many organizations treat them as interchangeable. It’s frustrating. This needs to change. DCR selection is not just a business decision—it’s a #strategic, #technical, and #compliance-driven choice that requires a blend of expertise across #dataprivacy, #analytics, and #security. While leadership plays a crucial role in setting the vision, the best outcomes come when cross-functional teams—those closest to the data, privacy regulations, and infrastructure—are actively involved in the decision-making process. DCR selection isn’t just another procurement exercise. It’s not about picking the biggest name or the most familiar vendor. It’s about understanding #privacyarchitectures, #interoperability, #security, #governancemodels, and use case alignment. I’d argue that a mid-level data privacy analyst or cloud engineer might make a better DCR decision than a C-suite executive with limited exposure to these intricacies. A wrong choice can jeopardize compliance, lead to inefficiencies, and sometimes expose sensitive data in ways you never anticipated. So, before you decide, ask the hard questions: 🔹 Is the DCR truly #neutral, or is it tied to a larger business interest (cloud, identity, media, or walled gardens)? 🔹 Does it allow #decentralized collaboration, or does it require centralizing my data? 🔹 Can I #enforce privacy controls, or can they be turned off entirely? 🔹 Does the provider become a #datacontroller under my local privacy regulations? 🔹 What Privacy-Enhancing Technologies (#PETs) are in place? 🔹 How fast can I generate #insights—instantly or after weeks of waiting? 🔹 Can I collaborate #globally, or am I restricted to a single region? DCRs are not plug-and-play solutions—they require a level of scrutiny that many in the industry are still not applying. So, let’s fix this. Let’s ensure the right people, with the right expertise, are leading DCR selection. Selecting the right one requires rigour, the right expertise at the table, and a clear understanding of how it aligns with business and privacy goals. Because in data privacy, the wrong decision isn’t just inefficient—it’s irreversible. Motivated by your piece Devon DeBlasio :) #DataCleanRooms #Privacy #Compliance #Martech #DigitalAdvertising

📗𝗗𝗮𝘁𝗮𝗯𝗿𝗶𝗰𝗸𝘀 𝗖𝗹𝗲𝗮𝗻 𝗥𝗼𝗼𝗺 : A Clean Room is a secure, privacy-centric environment that allows multiple organizations—or teams within the same organization—to collaborate on shared data without actually exposing the raw underlying datasets. It’s a concept born from the growing need for data collaboration under strict governance, compliance, and confidentiality requirements. 📗𝗥𝗲𝗾𝘂𝗶𝗿𝗲𝗺𝗲𝗻𝘁𝘀 𝗳𝗼𝗿 𝗖𝗹𝗲𝗮𝗻 𝗥𝗼𝗼𝗺𝘀 : To be eligible to use clean rooms, you must have:   • A workspace that is enabled for 𝘀𝗲𝗿𝘃𝗲𝗿𝗹𝗲𝘀𝘀 compute.   • A workspace that is enabled for 𝗨𝗻𝗶𝘁𝘆 𝗖𝗮𝘁𝗮𝗹𝗼𝗴.   • 𝗗𝗲𝗹𝘁𝗮 𝗦𝗵𝗮𝗿𝗶𝗻𝗴 enabled for your Unity Catalog metastore. 📗𝗖𝗼𝗻𝗰𝗲𝗽𝘁 𝗢𝘃𝗲𝗿𝘃𝗶𝗲𝘄 : A Databricks Clean Room is essentially a controlled data collaboration space. Imagine two companies—a retailer and a supplier—want to analyze joint performance metrics, optimize promotions, or run attribution models. Traditionally, sharing customer-level data would raise legal, privacy, and competitive concerns. 𝗔 𝗖𝗹𝗲𝗮𝗻 𝗥𝗼𝗼𝗺 𝘀𝗼𝗹𝘃𝗲𝘀 𝘁𝗵𝗶𝘀 𝗯𝘆 𝗲𝗻𝗮𝗯𝗹𝗶𝗻𝗴 𝗯𝗼𝘁𝗵 𝗽𝗮𝗿𝘁𝗶𝗲𝘀 𝘁𝗼:   1. Contribute their datasets securely (typically stored as Delta tables),   2. Run approved computations (queries, models, transformations) on the combined data, and   3. View only the aggregated, anonymized, or policy-approved outputs.   4. No one ever sees the other party’s raw data. 📗 𝗞𝗲𝘆 𝗣𝗿𝗶𝗻𝗰𝗶𝗽𝗹𝗲𝘀 : 𝗗𝗮𝘁𝗮 𝗣𝗿𝗶𝘃𝗮𝗰𝘆 & 𝗦𝗲𝗰𝘂𝗿𝗶𝘁𝘆   • Data never leaves the owner’s control.   • Access is strictly 𝗽𝗼𝗹𝗶𝗰𝘆-𝗱𝗿𝗶𝘃𝗲𝗻—even query-level rules can apply.   • Sensitive information can be masked, hashed, or aggregated automatically. 𝗚𝗼𝘃𝗲𝗿𝗻𝗮𝗻𝗰𝗲 & 𝗔𝘂𝗱𝗶𝘁𝗶𝗻𝗴   • Every action within a clean room is 𝘁𝗿𝗮𝗰𝗲𝗮𝗯𝗹𝗲 𝗮𝗻𝗱 𝗮𝘂𝗱𝗶𝘁𝗮𝗯𝗹𝗲.   • Permissions and controls are managed via Unity Catalog—the single source of truth for data governance in Databricks. 𝗖𝗼𝗹𝗹𝗮𝗯𝗼𝗿𝗮𝘁𝗶𝗼𝗻 𝗮𝘁 𝗦𝗰𝗮𝗹𝗲   • Partners can perform advanced analytics, ML model training, and reporting on shared insights—without compromising privacy.   • It’s built for cross-cloud and cross-organization use cases (leveraging Delta Sharing). #cleanrooms #deltalake #databricks #dataengineer #dataarchitect #governance

🚀 𝗘𝘅𝗽𝗹𝗼𝗿𝗶𝗻𝗴 𝗔𝘇𝘂𝗿𝗲 𝗗𝗮𝘁𝗮𝗯𝗿𝗶𝗰𝗸𝘀 𝗖𝗹𝗲𝗮𝗻 𝗥𝗼𝗼𝗺𝘀: 𝗔 𝗥𝗲𝘃𝗼𝗹𝘂𝘁𝗶𝗼𝗻 𝗶𝗻 𝗦𝗲𝗰𝘂𝗿𝗲 𝗗𝗮𝘁𝗮 𝗖𝗼𝗹𝗹𝗮𝗯𝗼𝗿𝗮𝘁𝗶𝗼𝗻! 🔒  💡 𝗪𝗵𝗮𝘁 𝗶𝘀 𝗔𝘇𝘂𝗿𝗲 𝗗𝗮𝘁𝗮𝗯𝗿𝗶𝗰𝗸𝘀 𝗖𝗹𝗲𝗮𝗻 𝗥𝗼𝗼𝗺𝘀? Imagine a secure environment where multiple organizations collaborate on sensitive enterprise data 𝘄𝗶𝘁𝗵𝗼𝘂𝘁 𝗲𝘃𝗲𝗿 𝘀𝗵𝗮𝗿𝗶𝗻𝗴 𝗿𝗮𝘄 𝗱𝗮𝘁𝗮 𝗱𝗶𝗿𝗲𝗰𝘁𝗹𝘆. Azure Databricks Clean Rooms makes this possible by combining 𝗗𝗲𝗹𝘁𝗮 𝗦𝗵𝗮𝗿𝗶𝗻𝗴 with 𝘀𝗲𝗿𝘃𝗲𝗿𝗹𝗲𝘀𝘀 𝗰𝗼𝗺𝗽𝘂𝘁𝗲, ensuring privacy and security every step of the way.  🔍 𝗛𝗼𝘄 𝗜𝘁 𝗪𝗼𝗿𝗸𝘀: - A 𝗰𝗲𝗻𝘁𝗿𝗮𝗹 𝗰𝗹𝗲𝗮𝗻 𝗿𝗼𝗼𝗺 acts as an isolated, ephemeral workspace managed by Databricks.   - Collaborators share metadata (column names, types) and run pre-approved notebooks in the central clean room.   - All computation happens securely within this no-trust environment—no direct access to raw data is allowed.   - Results can be temporarily saved as read-only output tables for further use.  💡 𝗧𝗵𝗲 𝗡𝗼-𝗧𝗿𝘂𝘀𝘁 𝗠𝗼𝗱𝗲𝗹   Azure Databricks Clean Rooms is built on a "no-trust" foundation:   - Collaborators *must approve* notebooks before execution.   - You can only execute notebooks created by your collaborator.   - Once created, the clean room is locked—no new collaborators can join.  🔐 𝗔𝗱𝗱𝗶𝘁𝗶𝗼𝗻𝗮𝗹 𝗦𝗮𝗳𝗲𝗴𝘂𝗮𝗿𝗱𝘀:   - Strict restrictions during the public preview, like a two-collaborator limit per clean room.   - Actions like renaming or adding collaborators are disallowed to maintain security.  ⚠️ 𝗟𝗶𝗺𝗶𝘁𝗮𝘁𝗶𝗼𝗻𝘀 (𝗣𝘂𝗯𝗹𝗶𝗰 𝗣𝗿𝗲𝘃𝗶𝗲𝘄): - No support for service credential Scala libraries.   - Resource quotas are enforced for Clean Room objects.  📌 𝗪𝗵𝗼 𝗦𝗵𝗼𝘂𝗹𝗱 𝗨𝘀𝗲 𝗧𝗵𝗶𝘀?   If you're in a data-driven industry where privacy is critical—think healthcare, finance, or retail—Azure Databricks Clean Rooms can enable secure partnerships and collaborations like never before.  🌟 𝗪𝗵𝘆 𝗜𝘁 𝗠𝗮𝘁𝘁𝗲𝗿𝘀:  Collaboration on sensitive data is often fraught with trust and compliance challenges. Clean Rooms removes these barriers, fostering innovation without compromising security.  💬 What are your thoughts on this cutting-edge feature? Would you use it for your collaborative projects? Let’s discuss!  #AzureDatabricks #DataCollaboration #DataPrivacy #CleanRooms #DeltaSharing #Innovation

#Sarus in #Azure #Confidential #CleanRoom enables multi-party data #privacy-safe collaboration at scale. The process of manually validating privacy policies for each data processing job is cumbersome, error-prone, and hard to scale. Sarus integrated with Azure Confidential Clean Rooms enhances automated and streamlined privacy validation, ensuring privacy-safe data collaboration at scale. A prime example of this capability in action is Sarus' partnership with EY to tackle financial crime. By combining transaction data from multiple banks, we’re able to uncover patterns that can help detect human trafficking. The challenge, however, lies in ensuring that this sensitive data is handled securely while still allowing researchers to explore different analytical models. Sarus and Azure Confidential Clean Rooms solve this problem by enabling dynamic, privacy-preserving data processing. Researchers can iterate on their models, receive anonymized results, and only pass on suspicious data to the regulator, ensuring full compliance without compromising the integrity of the analysis. the blog explains more in detail: https://lnkd.in/gQKEHWUi.

On the Security and Privacy of Federated Learning: A Survey with Attacks, Defenses, Frameworks, Applications, and Future Directions #Federated #Learning (#FL) is an emerging distributed machine learning paradigm enabling multiple clients to train a global model collaboratively without sharing their raw data. While FL enhances #data #privacy by design, it remains vulnerable to various #security and #privacy #threats. This survey provides a comprehensive overview of more than 200 papers regarding the state-of-the-art attacks and defense mechanisms developed to address these challenges, categorizing them into security-enhancing and #privacy - #preserving #techniques. Security-enhancing methods aim to improve FL robustness against malicious behaviors such as byzantine attacks, poi soning, and #Sybil #attacks. At the same time, privacy-preserving techniques focus on protecting sensitive data through #cryptographic approaches, differential privacy, and secure aggregation. We critically analyze the strengths and limitations of existing methods, highlight the trade-offs between #privacy, #security, and model performance, and discuss the implications of non-IID #data distributions on the effectiveness of these defenses. Furthermore, we identify open research challenges and future directions, including the need for scalable, adaptive, and energy-efficient solutions operating in dynamic and heterogeneous FL environments. Our survey aims to #guide #researchers and practitioners in developing robust and privacy-preserving FL systems, fostering advancements safeguarding collaborative learning frameworks’ #integrity and #confidentiality. Centro de Investigación de Ciberseguridad IoT - IIoT

Solving the Linkage Problem is the missing piece in many Privacy Enhancing Technologies (PETs). PETs are evolving, but many still miss addressing the biggest challenge: linking data across partners without exposing identities. PETs like federated learning, differential privacy, fully homomorphic encryption, and synthetic data have strengths, but they focus on already integrated data and don’t solve the linkage problem. And data clean rooms, while being touted as a privacy solution, still require centralizing data with a third party. True privacy-first data collaboration requires new approaches—ones that don’t sacrifice accuracy for security. The future belongs to solutions that can link datasets without exposing sensitive information.

One of the most common questions I hear in collaborative analytics conversations is: how do you actually run joint analysis across organizational boundaries without sharing raw data? This practical walkthrough from Nawaz Dhandala covers exactly that — standing up AWS Clean Rooms end to end, from collaboration setup and membership configuration to defining analysis rules and running protected queries. As data collaboration across organizations becomes more common, Clean Rooms fills a genuine gap between “just share the raw data” (risky) and “build custom secure compute infrastructure” (expensive). Worth a read if you’re thinking about how to enable cross-organizational analytics without compromising data privacy. https://lnkd.in/gWFHV5Cy #AWS #Analytics #DataPrivacy #CleanRooms #DataCollaboration