0x30D - Teknik - Panel nsec 2026

0x30D - Teknik - Panel nsec 2026

by PolySécure Podcast

Applied Sciences & Engineering Computing & Cybersecurity UComputing and Information TechnologyURComputer security
🎙 PolySécure Podcast 👥 539 📅 June 18, 2026 ⏱ 51 min 👁 25 📄 expert opinion 🧭 2026-08-16
Available in: English (current) Français

Keywords

MythosLLMvulnerability researchsupply chainred team

Summary

The panel discusses the impact of AI on cybersecurity, focusing on the Mythos model. They argue that Mythos is not fundamentally more dangerous than other models, attributing its hype to marketing and a catchy name. The vulnerability in BSD was already known, and the model’s success is partly due to training data. The discussion highlights the importance of the attacker’s skill level, with many attackers being ‘brain dead’ and wasting tokens. The cost of AI-driven attacks is a barrier, and finding vulnerabilities doesn’t guarantee exploitability or faster patching. They note that alert volumes haven’t spiked, contradicting fears of mass AI-driven attacks. The panel also discusses how LLMs introduce new vulnerabilities, such as supply chain attacks and leaked secrets. They emphasize the need for better network visibility and the human bandwidth to review AI-generated code. The conversation touches on the criminal ecosystem, noting that criminals are opportunistic and may not adopt expensive AI tools. They conclude with open questions about technological debt and the future of vulnerability research.

167 words

Critical Evaluation

Value of the Information & Strength of the Argument

The value of the information is high, as it provides a balanced, expert perspective on AI in cybersecurity, countering hype with practical experience. The argumentation is solid, with panelists supporting their points with real-world examples and analogies. They effectively debunk myths about AI autonomy and emphasize the role of human expertise.

Scientific Rigor, Source Quality, Title Accuracy

The discussion is rigorous, with panelists drawing on their professional experience. No formal sources are cited, but references to known researchers like Nicolas Carlini and specific incidents add credibility. The title accurately reflects the content, and the panel’s expertise lends authority.

107 words

Title / Content Match

Title accurately reflects the content: a technical panel discussion at NSEC 2026.

Quality & Reliability

8/10

Panel of five experienced cybersecurity professionals discussing AI-driven vulnerability research, with balanced perspectives and practical insights. No formal citations but grounded in professional experience.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Cited Sources

  • This is How They Tell Me the World Ends — Mentioned by panelist as a book by Nicole Perlroth, relevant to AI and zero-days.
  • Nicolas Carlini's research — Referenced for prior work on LLM vulnerability discovery.

Concurring Sources

  • Nicolas Carlini's research — Panelists agree with his findings on LLM vulnerabilities.

Contribution & Novelties

The panel provides a nuanced, practitioner-based perspective on AI in cybersecurity, challenging the hype around models like Mythos. It emphasizes the importance of human expertise and the practical limitations of AI-driven attacks.

Pour aller plus loin :

  • LLM for Vulnerability Discovery — Relevant research on using LLMs for vulnerability discovery.
  • Supply Chain Security — Overview of supply chain attacks.
  • NVD — National Vulnerability Database, relevant to patching challenges.

68 words

Radar Profile

The radar shows high scores in information quantity, quality, and technical level, with a slightly lower but still strong reliability score. This indicates a well-rounded, expert discussion with practical insights.

Reliability 8/10