Pii
PulseAugur coverage of Pii — every cluster mentioning Pii across labs, papers, and developer communities, ranked by signal.
7 day(s) with sentiment data
-
LLM governance engine adds RAGAS faithfulness scoring to combat hallucinations
A developer has enhanced an LLM governance engine by integrating RAGAS faithfulness scoring, which measures how well a model's response aligns with provided context. This new feature complements the existing PII firewal…
-
AI's role in Data Loss Prevention: Moving beyond detection to judgment
Traditional Data Loss Prevention (DLP) tools focus on detection through pattern matching, but this approach has limitations. A recent study found that over 80% of findings from regex and PII-focused models were false po…
-
LLM guardrails enhance AI safety by validating inputs and outputs
LLM guardrails are being developed to provide an independent validation layer for AI systems. These guardrails aim to intercept unsafe inputs and malformed outputs before they enter production environments. Specifically…
-
5 techniques to anonymize PII in LLM pipelines
Protecting personally identifiable information (PII) within large language model (LLM) pipelines is a critical but often overlooked aspect of AI development. Data used for training, fine-tuning, retrieval-augmented gene…
-
AI agents struggle to differentiate facts from guesses due to data catalog flaws
A critical flaw exists in AI data agents where they cannot distinguish between factual information and model-generated inferences due to data architecture failures. This issue stems from data catalogs flattening distinc…
-
Prompt injection attacks on LLMs have two types, with indirect injection posing a greater threat
Prompt injection attacks on large language models (LLMs) are more complex than commonly assumed, with two distinct types. Direct injection involves malicious instructions from the user, which most systems are designed t…
-
AWS offers best practices for Bedrock Guardrails in code generation
AWS is providing best practices for implementing Amazon Bedrock Guardrails in code generation workflows. These guardrails are crucial for detecting and filtering unsafe code patterns, preventing prompt attacks, and reda…
-
Platform engineers unexpectedly become AI guardrail guardians
Platform engineers are increasingly finding themselves responsible for AI guardrails, a role they did not anticipate. This responsibility is split between feature-side developers integrating LLMs into user-facing produc…
-
Node.js LLM Security Toolkit Launched by Resk-Security
Resk-Security has released resk-llm-ts, a new TypeScript toolkit designed to protect Node.js Large Language Model (LLM) applications from security threats. The toolkit offers 11 threat detectors to address issues like p…
-
New CBOM spec aims to bring SBOM-like provenance to LLM prompts
The author proposes Context Bill of Materials (CBOM), an open specification and SDK, to address the lack of auditable provenance for data fed into large language models. Similar to Software Bill of Materials (SBOM) used…
-
India's DPDP Act intensifies data privacy compliance risks for offshore engineering centers
Companies with offshore engineering centers in India face increased scrutiny and potential penalties due to the enforcement of the Digital Personal Data Protection (DPDP) Act. This legislation requires local entities to…
-
LLM Guardrails: Protecting AI Apps from Prompt Injection and Data Leaks
LLM guardrails are essential for securing AI applications by acting as a protective layer between user input and the language model. These guardrails help prevent prompt injection attacks, where malicious instructions o…
-
LLM evaluations must weigh failure severity, not just pass rates
A recent LLM deployment experienced a PII leak, where an agent accidentally included a customer's account ID and partial billing address in a support response. This incident occurred despite the evaluation dashboard sho…
-
Trump's National Design Studio open-sources Rampart AI model
Donald Trump's National Design Studio has open-sourced Rampart, a MiniLM-based AI model. This release grants global access to the source code, allowing users to understand how the AI stores detected Personally Identifia…
-
New testbed LACUNA evaluates LLM unlearning precision at parameter level
Researchers have introduced LACUNA, a novel testbed designed to evaluate the precision of unlearning methods for large language models (LLMs). Current unlearning benchmarks focus solely on output-level performance, fail…
-
New middleware filters PII from AI agent payment requests
Researchers have developed a new open-source middleware called presidio-hardened-x402 designed to protect personally identifiable information (PII) in AI agent payment requests. This system filters metadata, such as res…
-
ComplianceGate system routes LLM inferences for regulated industries
Researchers have developed ComplianceGate, a novel architecture for routing large language model (LLM) inferences in regulated industries. This system uses a pre-inference classifier to evaluate query complexity and dat…
-
AI Gateways Enhance Data Security for ChatGPT and Claude Usage
Organizations are implementing Data Loss Prevention (DLP) strategies to prevent sensitive data from being exposed when using third-party Large Language Models (LLMs) like ChatGPT and Claude. An AI gateway, such as Bifro…
-
Automated redaction secures sensitive data for LLM use
To mitigate security and compliance risks when using LLMs like those from OpenAI, Anthropic, and Google, sensitive data must be redacted before being sent in prompts. This involves automated inline prompt redaction, whi…
-
AI security: Multi-layered approach to prevent sensitive data leakage
Organizations must implement a multi-layered security strategy to prevent sensitive data from being sent to third-party AI tools. This involves identifying and classifying data such as PII, PHI, secrets, and intellectua…