Generative AI is poised to revolutionize vulnerability discovery in critical infrastructure, but will it actually fix the problem, or just shift the burden?

The recent AI Cybersecurity Challenge (AIxCC), a two-year competition sponsored by the US Defense Advanced Research Projects Agency (DARPA) and Advanced Research Projects Agency for Health (ARPA-H), crowned winners whose AI systems autonomously discovered and patched zero-day flaws in real-world code.

Now, with models potentially going open-source, the implications for defenders, attackers and policymakers are seismic.

In this episode, we sat down with Taesoo Kim, the leader of Team Atlanta, the AIxCC winning team, and Andrew Carney, program manager for the AIxCC at DARPA and ARPA-H.

In the interview (13.56), they discuss why the commercialization of GenAI-powered vulnerability scanning tools could be just around the corner and how "self-healing infrastructure" might soon become a reality.

Can AI Solve the Vulnerability Problem in Critical Infrastructure?

W dzisiejszym odcinku Jan Jęcz opowiada o chińsko-amerykańskich relacjach gospodarczych i handlowych, o możliwym porozumieniu w sprawie ceł, o wymuszonej sprzedaży TikToka oraz zamknięciu się Chin na amerykańskie czipy.

Rozmowę prowadzi Karol Tokarczyk.

Zapraszamy!

Sprzedaż TikToka, zakaz dla Nvidii | Z dystansu

Abopeg

AVE|CALADRA\gloria\liturgy\

Medium Article: https://medium.com/@jsmith0475/ai-that-thinks-backward-the-rise-of-defensive-intelligence-c0260765a2ed
The academic paper, by Dr. Jerry A. Smith, introduces "Defensive Intelligence" as a new architectural principle for agentic AI, arguing that inversion reasoning—explicitly modeling and avoiding failure modes—significantly improves system robustness over traditional goal-oriented methods. It proposes four technical patterns, such as Adversarial Attention Heads and Failure Mode Memory, that embed this defensive mindset directly into transformer architectures, claiming up to a forty percent reduction in task failures. Beyond implementation, the source explores the profound implications of this failure-aware AI, addressing the cognitive asymmetry between defensive AI and optimism-biased humans, the sociological risks of concentrating "negative knowledge" among elite actors, and the ethical challenges of prioritizing which failures the AI should avoid. Ultimately, the work suggests that this type of defensive reasoning may result in an intelligence that is fundamentally more cautious and alien than human cognition.

AI That Thinks Backward: The Rise of Defensive Intelligence

🚀 This week on the Tech ONTAP Podcast: NetApp Docs goes AI!

We sit down with Aksel Davis, Grant Glass, Jen Kaufman, and Adam Newton to explore how NetApp is transforming documentation—from static PDFs to a living, searchable, AI-powered ecosystem.

💡 Topics we cover:
	•	The shift from closed XML systems to GitHub + open standards
	•	How generative AI powers Doc the Owl, our chatbot trained on 120K+ NetApp documents
	•	Making documentation more accurate, agile, and accessible in 9 languages
	•	Future possibilities: AI-driven authoring tools, in-product assistance, and cross-dataset intelligence

Whether you’re a NetApp customer, partner, or just curious about how AI reshapes technical documentation, this episode is packed with insights.

Ep 406 - AI Powered NetApp Docs

Medium Article: https://medium.com/@jsmith0475/why-current-nil-valuations-fail-and-how-multi-agent-ai-fixes-them-f1652a0e887c
The article, by Dr. Jerry A. Smith, describes VALORE, a novel multi-agent artificial intelligence system designed to accurately value a collegiate athlete's Name, Image, and Likeness (NIL) influence, correcting for the failures of current surface-level metrics. This system employs seven specialized thinking transformer models—such as a Social Media Analysis Agent and a Psychological Profile Agent—that coordinate through goal-oriented consensus mechanisms to integrate diverse factors like behavioral science, economic data, and athletic performance. The research emphasizes that VALORE models crucial human elements like parasocial relationships and authenticity to predict true marketing value, ensuring the system maintains high prediction accuracy, transparent coordination, and ethical oversight through proactive bias detection. Ultimately, VALORE seeks to create more equitable and efficient markets by benefiting athletes, brands, and universities through enhanced decision support and compliance.

Why Current NIL Valuations Fail — and How Multi-Agent AI Fixes Them

Medium Article: https://medium.com/@jsmith0475/the-53-problem-what-traditional-nil-valuations-miss-2ab9fd53d595
The article "The 53% Problem: Cultural Factors in NIL Valuation," by Dr. Jerry A. Smith, argues that traditional Name, Image, and Likeness (NIL) athlete valuation models are fundamentally flawed because they fail to account for cultural factors that contribute to 53% of the variance in market value. The core premise is that characteristics such as gender, race, institutional prestige, and geographic location do not combine additively but rather interact through multiplication, leading to dramatically compounded disadvantages for some athletes. The text proposes using mathematical frameworks, specifically differential equations, as reasoning anchors for multi-agent Artificial Intelligence (AI) systems to model these complex, multiplicative cultural dynamics consistently and accurately. This approach is intended to expose systematic inequities, such as the significant financial penalties faced by international or female athletes, and to provide data-driven strategic guidance for interventions. The source also discusses the ethical challenges and need for empirical validation of these mathematically anchored AI models before their superiority can be confirmed.

The 53% Problem: What Traditional NIL Valuations Miss

Medium Article: https://medium.com/@jsmith0475/we-solved-ais-reproducibility-crisis-by-treating-it-like-a-physics-problem-8936aed52923
The article "Cognitive Anchoring," by Dr. Jerry A. Smith, details a novel solution to the reproducibility crisis in large language models (LLMs) by treating the issue as a physics coordination problem. The core proposal, cognitive anchoring, uses principles from gauge theory to synchronize the attention heads within transformer models, which otherwise drift and produce inconsistent reasoning paths. The authors introduce four specific anchoring mechanisms—symbolic, temporal, spatial, and symmetry—to constrain representational degrees of freedom without sacrificing logical content, leading to a 38% improvement in symbolic consistency during complex tasks like discovering field equations. The framework is presented as a mechanistic alternative to prompt engineering and is demonstrated to generalize across scientific discovery and behavioral science applications, such as modeling complex cultural multipliers in athletic valuation. Ultimately, the paper establishes anchoring as a foundational protocol for achieving stable and reliable inference in AI reasoning systems.

We Solved AI's Reproducibility Crisis by Treating It Like a Physics Problem

Medium Article: https://medium.com/@jsmith0475/why-your-ai-agents-keep-failing-and-what-synthetic-intelligence-can-do-about-it-416f035266bc
"Synthetic Intelligence: Why AI Agents Fail and What Comes Next," by Dr. Jerry A. Smith, details the widespread failure of current enterprise AI agents, citing failure rates as high as 95% for pilots and high operational costs due to unsustainable energy consumption. The author argues that transformer-based AI is fundamentally limited because it can only respond and simulate intelligence, lacking the capacity for genuine autonomy, intrinsic motivation, and continuous learning required for complex business tasks. As an alternative, the text introduces Synthetic Intelligence (SI), an architecture based on neuromorphic computing and Psi-Theory, which replicates biological brain functions to create non-biological intelligence that is vastly more energy-efficient and capable of genuine adaptive decision-making. The author strongly advises a hybrid strategy where businesses continue using existing reactive AI for simple tasks while immediately investing in SI to gain a competitive advantage in building truly autonomous systems.

Why Your AI Agents Keep Failing - And What Synthetic Intelligence Can Do About It

In-Depth Explorations of Neuroscience-Inspired Architectures Revolutionizing AI.

Each week, the Tech ONTAP Podcast dives into all-things NetApp, including storage, public & private cloud, and much more. The team will also be interviewing subject-matter experts from across the industry and detailing various best practices for getting the most out of your datacenter.

a/k/a CHRIS CYPHER
[FUTURE_WITCH\TRICK]
author of 'LANTERNGLASS'
and
'an UNASSEMBLED VERSE'
verseweld.substack.com

Podcasty Polityki Insight skupiają się na najważniejszych wydarzeniach politycznych i gospodarczych, analizujemy krajowe, europejskie i globalne trendy.

Interested in the latest news from the Infosecurity sector and want to hear from the industry’s leading cybersecurity experts?

Listen to our monthly podcasts to hear our award-winning editorial team discuss the latest trends and issues in information security. The team is regularly joined by cybersecurity experts who take a deep dive into the most important issues of the day and provide you with solutions and practices to help your business stay secure.

New tracks tagged #LLM