RadioCSIRT English Version - Special Edition Claude IA - Cyber Security Podcast, (Show 72)

🎧🎙️Large Language Models & Cybersecurity Claude 4.6, Project Glasswing & Claude Mythos PreviewWelcome to this special edition of RadioCSIRT ⚡️🤖 Claude 4.6 Family — A Cybersecurity Perspective Claude Opus 4.6, Sonnet 4.6 and Haiku 4.5 share a context window of up to one million tokens, multimodal text/image support and extended thinking capabilities. Direct implications for SOC teams: complete code repository analysis, massive IOC correlation, attack chain reconstruction — but also a significant reduction in the entry barrier for producing high-quality offensive artifacts.🔓 Economic Asymmetry of the Threat At five dollars per million tokens, Claude Opus 4.6 makes expert-level analytical reasoning accessible to a broad range of actors that previously required costly human expertise. The window between CVE publication and exploit availability is compressing. LLM-generated phishing lures no longer display the linguistic markers traditionally used for detection.🔬 Project Glasswing — Restricted Access Anthropic launched Claude Mythos Preview under an access framework limited to approximately forty partner organizations (Microsoft, Google, Amazon confirmed), by invitation only, following prior consultation with US authorities. The European Commission publicly endorsed this restriction.⚠️ Claude Mythos Preview — Documented Capabilities The model can autonomously identify and exploit flaws across all major operating systems and web browsers, and construct sophisticated payloads and exploits in real time at low cost. On April 7, 2026, Treasury Secretary Scott Bessent and Jerome Powell convened an emergency meeting with the CEOs of major US banks (Bank of America, Citigroup, Goldman Sachs, Morgan Stanley, Wells Fargo) — the first meeting of this level motivated by the capabilities of a single AI model.🎯 Identified Risk Vectors Six dimensions covered in this episode: zero-day exploitation, systemic SIFI risk, algorithmic convergence, DeFi/smart contract exposure, customer data exfiltration, impact on cyber-insurance portfolios.⚖️ Regulatory and Legal Context Anthropic is in active litigation with the Pentagon, which classified the organization as a supply chain risk. The forty Glasswing partners constitute a new indirect attack surface. The AI Act, DORA and ENISA guidelines create a compliance framework that applies now to LLM deployments in high-risk contexts.🛡️ Documented Defensive Use Cases Automatic generation of YARA/Sigma rules, alert enrichment, large-scale forensic analysis, assisted threat modeling, adversary simulation — the same capabilities serve both sides. LLMs remain analytical augmentation tools: human verification on high-impact operational outputs remains mandatory.🔗 SourcesProject Glasswing — Anthropic: https://www.anthropic.com/glasswingClaude Models — Overview and Pricing: https://platform.claude.com/docs/en/about-claude/models/overview📩 Contact: [email protected] ⚡️ We don't think, we patch!#RadioCSIRT #Cybersecurity #LLM #Claude #Anthropic #Glasswing #Mythos #ThreatIntelligence #SOC #CERT #CISO #AI #CyberSecurity #ProjectGlasswing #ZeroDay #SIFI