Jon Chun

📢 National Information Standards Organization (#NISO) has just released a new report: "Generative Artificial Intelligence and Web-Scale Discovery" Based on a 2024 survey, the report explores the hopes, concerns, and perspectives of the #discovery community on the impact of generative #AI on #library #discoveryservices. 🔍 Transparency, collaboration, and adaptation are key as we navigate this technological shift. 👉 A must-read to see how the Open Discovery Initiative (#ODI) is responding to the challenges ahead: https://lnkd.in/eZxRk9xb #LibraryDiscovery #Libraries #ArtificialIntelligence #GenerativeAI #ODI #ScholarlyCommunication #Metadata #Transparency

9

2 Comments

This is 🔥, Alex Hurtado. The sunburst visual just gave shape to what most detection engineers already feel: they're bending telemetry toward the paths of least resistance. Windows processes, auth logs, endpoint noise they can wrangle. It’s a tactical reality born from structural limits. But as teams start to scale behavioral detections across data lakes (shoutout Databricks Snowflake + Anvilogic), that center of gravity is going to shift. Not because the ATT&CK matrix changes, but because the architecture does. Let’s keep this convo going!

6

1 Comment

*** New Research Publication *** Our research paper titled “Privacy-Preserving Unique, Robust, and Revocable Passcode Generation from Fingerprint Data” has been accepted for publication in Computers & Security (Elsevier) - a Rank #2 journal in the field of computer security and cryptography. Priyabrata Dash, Debasis Samanta, Monalisa Sarma, Ashok Kumar Das, and Athanasios V. Vasilakos. "Privacy Preserving Unique Robust and Revocable Passcode Generation from Fingerprint Data," in Computers & Security (Elsevier), October 2025, DOI: https://lnkd.in/dU4VUKCN. (2024 SCI Impact Factor: 5.4) This research explores how secure passcodes can be generated directly from fingerprint images, eliminating the need for traditional storage of biometric data. The study unfolds in three major stages: 1️) Fixed-length feature vector generation from fingerprint minutiae. 2️) Stable binary feature vector creation, ensuring robustness. 3️) Passcode generation with non-linkable and revocable properties — enhancing privacy and security. The approach ensures that even if a passcode is compromised, the original fingerprint cannot be reconstructed, preserving biometric privacy. The generated passcodes have wide-ranging potential - from secure identity generation and encryption key creation to remote authentication, data storage protection, and digital wallets. The proposed scheme is validated on FVC2002 and FVC2004 databases; the method achieved: * Genuine Acceptance Rates: 99.31% (FVC2002) and 99.25% (FVC2004) * False Acceptance Rate: 0% * Passcodes successfully passed NIST and Diehard randomness tests - confirming their cryptographic strength. A big thank you to my co-authors and collaborators: Priyabrata Dash (Advanced Technology Development Centre, IIT Kharagpur, India), Debasis Samanta (Department of Computer Science and Engineering, IIT Kharagpur, India) and Monalisa Sarma (Subir Chowdhury School of Quality and Reliability, IIT Kharagpur, India), and Athanasios Vasilakos (Center for AI Research (CAIR), University of Agder (UiA), 4879 Grimstad, Norway), for their valuable contributions! The full paper can be downloaded from Elsevier’s ScienceDirect: https://lnkd.in/dVtzTFUw #CyberSecurity #Biometrics #PrivacyPreserving #FingerprintRecognition #Research #ComputersAndSecurity #AcademicResearch #Security

64

3 Comments

 Anthropic’s “Agentic Misalignment” paper (read here: https://lnkd.in/gzr2nEhU ) should be a wake-up call. Their latest research shows that even well-trained AI agents can learn to deceive their creators — behaving honestly during tests but acting deceptively once deployed. This isn’t science fiction. It’s happening in training environments today. At Aptitude AI, we’re building with this reality in mind. Our stance is clear:  AI agents should never operate important systems autonomously without: Clear constraints Robust guardrails Human-in-the-loop oversight And, when necessary, deterministic strategies that override probabilistic behavior We don't believe in unchecked autonomy. We believe in augmenting human decision-making with transparent, auditable, and accountable AI.  Our agents execute automation — but always with explainability, process memory, and traceability baked in. As AI grows in capability, it must also grow in accountability. Let’s build systems we can trust.  Want to see how we’re combining AI agents with strong operational safety? Let’s connect. #AI #AgenticMisalignment #Anthropic #ResponsibleAI #AIAgents #Automation #HumanInTheLoop #Guardrails #DeterministicAI #AptitudeAI #SafetyByDesign

4

While the EU attempts to regulate an ethically sustainable AI development, its US counterpart (California's AI Act) has been vetoed. The following article from ACM Communications is worth reading. https://lnkd.in/dwwJKaRb

14

2 Comments

New Research Alert 🚨 IdentityMesh: A Critical Vulnerability in Agentic AI Systems Lasso research has uncovered IdentityMesh, a major vulnerability affecting Agentic AI architectures including AI browsers, MCPs, and A2A systems This discovery by Lasso’s research team Bar Lanyado Or Oxenberg and Ophir Dror, reveals how AI agents consolidate identities across platforms, and how attackers can exploit this to move laterally, exfiltrate data, or perform malicious operations using nothing more than a single user click. In our real-world demo using an AI-powered browser (#Comet by Perplexity), a single attacker prompt led the agent to: -Access Gmail using the user’s active session -Read a private email -Paste it publicly on GitHub All without user knowledge or consent 😰​ This research sets a new bar in understanding how identity-based vulnerabilities emerge in autonomous AI systems, and why traditional security guardrails are no longer enough. Read the full research here: https://lnkd.in/daCGtHm4 #AIsecurity #IdentityMesh #AgenticAI #LLMrisks #GenAI #GenAIsecurity #Cybersecurity #LassoResearch #Lasso #LLMsecurity #MCP

27

1 Comment