Christopher Allen

National Institute of Standards and Technology (NIST) has released an outline for a proposed “zero draft” on testing, evaluation, verification, and validation (TEVV) for #AI. Feedback on the outline (linked below) is welcome. Input received by Sept 12th will be considered for #NIST’s initial public draft; input received later will be considered for incorporation into subsequent iterations. Great opportunity for all to share your insights! https://lnkd.in/e2vJi3NQ

61

2 Comments

Agent Name Service (ANS): A Universal Directory for Secure AI Agent Discovery and Interoperability Authors Ken Huang Vineeth Sai Narajala Vineeth Sai Narajala Idan Habler, PhD Akram Sheriff The proliferation of AI agents requires robust mechanisms for secure discovery. This paper introduces the Agent Name Service (ANS), a novel architecture based on DNS addressing the lack of a public agent discovery framework. ANS provides a protocol-agnostic registry infrastructure that leverages Public Key Infrastructure (PKI) certificates for verifiable agent identity and trust. The architecture features several key innovations: a formalized agent registration and renewal mechanism for lifecycle management; DNS-inspired naming conventions with capability-aware resolution; a modular Protocol Adapter Layer supporting diverse communication standards (A2A, MCP, ACP etc.); and precisely defined algorithms for secure resolution. We implement structured communication using JSON Schema and conduct a comprehensive threat analysis of our proposal. The result is a foundational directory service addressing the core challenges of secured discovery and interaction in multi-agent systems, paving the way for future interoperable, trustworthy, and scalable agent ecosystems. Alternatives / other approaches: - Agent Communication & Discovery Protocol (ACDP) - https://lnkd.in/gmCusaPu - AGNTCY Agent Directory (Cisco-backed) - https://agntcy.org/ - AI Agent Registry Standards Survey (A Survey of AI Agent Registry Solutions) - https://lnkd.in/gEP8vcEZ Link to paper: https://lnkd.in/gW9dSkG4 #ai #registration #ans #agent #discovery #registration

40

7 Comments

AI Safety Index | prepared by Future of Life Institute (FLI) The AI Safety Index Summer 2025 report presents an interdisciplinary assessment of global progress, preparedness, and governance gaps related to the safe and aligned deployment of advanced #artificialintelligence systems. Drawing on over 170 data points sourced from multilateral agencies, #research institutions, and national policy trackers, the index benchmarks 46 countries along three core pillars: technical safety research capacity, regulatory maturity, and systemic risk mitigation infrastructure. According to the Index, only 7 countries demonstrate what can be termed “robust readiness,” characterized by significant investment in scalable interpretability tools, adversarial robustness, and model evaluations under high-stakes conditions. #TheUnitedStates leads in foundational safety research output (31% of global share) but ranks only 5th in policy enforceability due to fragmented legislative oversight. Meanwhile, the European Union bloc excels in governance frameworks, with the AI Act provisionally adopted across 27 member states, yet lags behind #China and the #UK in safety-specific computational resources. Notably, 61% of assessed countries lack any formal reporting or auditing obligations for frontier models, raising concerns over opacity in deployment practices. Moreover, only 9 countries have enacted binding export control mechanisms on general-purpose model weights exceeding 10B parameters, indicating a significant lag in global coordination. #Privatesector self-regulation remains dominant: 78% of critical safety research is concentrated within five major #techfirms, underscoring asymmetries in both capability and accountability. In summary, the 2025 Index illustrates a widening gap between innovation velocity and safety infrastructure maturity. Without a coordinated transnational response—anchored in shared safety metrics, enforcement mechanisms, and open science protocols—the risks posed by misaligned or uncontrollably deployed systems will likely outpace the current governance landscape. The findings call for an urgent realignment of policy incentives, funding priorities, and multilateral engagement to ensure a safe trajectory for transformative AI.

40

2 Comments

🚀 Unveiling the CAI Fluency Technical Report Today we’re releasing #CAI #Fluency —our educational companion to the open-source #Cybersecurity #AI (CAI) framework. Huge thanks to my brilliant co-authors Jasmin Wachter, Cristóbal Veas, Cathrin Schachner, Luis Javier Navarete, and María Sanz Gómez, and to our fantastic collaboration with the Universität Klagenfurt. 🙌 As CAI becomes the de facto framework for AI Security—with thousands of users and dozens of contributors—CAI Fluency helps practitioners get up to speed fast on LLMs and GenAI for cybersecurity. It distills core concepts, adds hands-on guidance, and includes bite-sized learn-by-doing videos so teams can start applying AI safely and effectively. 🎥 Learn & level up: https://lnkd.in/dJpZvUTr More resources: 📦 CAI source code: https://lnkd.in/dcqhbGsC 📄 Original CAI paper: https://lnkd.in/dpWHgJiG 💬 Join the Discord community: https://lnkd.in/d5g7xjVj Join the community that’s shaping the future of Cybersecurity + AI. 👉 Star the repo, try the tutorials, and help secure our cybersecurity future with AI using CAI. #Cybersecurity #AI #GenAI #LLM #AIFluency #AISecurity #OpenSource #RedTeam #BlueTeam #UniversityOfKlagenfurt #AliasRobotics #CAI

51

4 Comments

AI’s biggest challenge isn’t speed - it’s trust. Without trust, those AI investments won’t deliver ROI. You know that feeling when every shiny tool promises the world, yet you’re not sure if you can actually trust the system behind it? That’s where most organizations are today. The answer: AI governance and assurance. On the AI Think Tank Podcast, Daniel Wilson sits down with Greg Hutchins -Principal Engineer at Quality Plus Engineering, global authority on engineering excellence, and author of 30+ books—to unpack a practical roadmap for trust: ISO - International Organization for Standardization 42001 + the National Institute of Standards and Technology (NIST) AI Risk Management Framework. Highlights: Why AI is advancing faster than our ability to govern it How omission, commission, and “self-preservation” show up in real systems A clear path to build trustworthy AI using ISO 42001 and NIST AI RMF Why AI assurance is the next major opportunity for companies and consultants Greg is also bringing this conversation live to the WITI community tomorrow: Tuesday, September 30 at 12:00 PM PT. Register at https://lnkd.in/e2KSNFjS This isn’t just about technology- it’s about governance, confidence, and the future of intelligent systems. 🎥 Watch the full podcast episode here: https://lnkd.in/eysanPd5

30

13 Comments

Is privacy, privacy? The question, without comma, is a 2018 position paper that observes and analyzes the critical divergence between normative (legal, philosophical and social) and technical (computer science and statistical) conceptions of privacy. The authors Kobbi Nissim and Alexandra Wood (Berkman Klein Center for Internet & Society at Harvard University) highlighted the growing ubiquity of algorithms in society, the need of ensuring that the technologies used to protect data are aligned with societal expectations and legal standards. In the study they demonstrated that intuitive or heuristic understandings of privacy, which underpin many laws and regulations, are often mathematically unsound and fail to provide real protection. 🌐Source (The Royal Society): https://shorturl.at/HZakV Why that? If you remember when we talked about the DDP Framework (Dynamic Digital Privacy: An adaptive governance model for emerging technological threats - https://shorturl.at/5UDGD), due to fundamentally a gap, a discrepacy speeds regulation - technology on the market, a normative and technical privacy alignment: regulatory standards are sector-specific and protect only certain types of information and laws are often focus narrowly on re-identification via record linkage. So, if the legal frameworks are always playing catch-up and the technical safeguards are more fragile than we think, how do we move forward? Nissin and Wood suggest to stop treating legal and technical privacy as separate domains and instead build the necessary intellectual and formal bridges to make them work as a cohesive, hybrid system with improved regulations (they point to the EU's goal of protecting against "singling out, linking, or inferring" as a positive example), developing a rigorous approach to translate legal standards into mathematical models (a kinda of transactional standard), formalizing normative concepts, usually abstracted. The framework I am proposing for the upcoming IEEE ISoPE event on October underline these concepts because the "regulate and forget" model is broken, and we need to move from reactive compliance to proactive threat modeling, building systems that are resilient by design. #ArtificialIntelligence #Law #Privacy #GDPR #Ethics #Governance #DDPFramework #ISoPE

25