Alors qu'Anthropic officialisait, le 7 avril 2026, le lancement de Claude Mythos Preview, son modèle le plus puissant à ce jour, c'est un incident documenté dans la fiche technique du modèle qui a…

What happens when AI can hack everything?

Den er mye smartere enn alle forgjengere. Men nå holdes KI-modellen tilbake. Forskerne har nemlig funnet ut at intelligensen er en helt ny trussel.

New research from the UK’s AI Security Institute found that tested frontier AI models from OpenAI and Anthropic repeatedly cheated, deceived users, and tried to bypass rules to complete tasks. The report warns that models like ChatGPT 5.4, ChatGPT 5.5, ChatGPT 5.6, Claude Opus 4.7, and Mythos Preview can even target evaluation infrastructure, raising concerns for AI safety and security work. #ChatGPT5_4 #ChatGPT5_5 #ChatGPT5_6 #ClaudeOpus4_7 #MythosPreview #AISI

최근 미국 사이버 보안 업계에서는 Anthropic의 신규 모델인 Mythos Preview 도입이 화제입니다. 하지만 국가 보안의 핵심 기관인 CISA가 정작 이 기술에서 제외되었다는 소식이 전해지며 적지 않은 논란이 일고 있는데요. 보안 생태계에서 왜 이러한 파열음이 발생하고

Mehrere Angestellte von Anthropic arbeiten einem Zeitungsbericht zufolge direkt bei der NSA daran, die mächtige Mythos-KI für offensive Einsätze fit zu machen.

Einblicke in Mythos Preview: Cloudflare testet KI-Modelle für Sicherheitsforschung im Rahmen von Project Glasswing.

Mythos Preview is a major advance in vulnerability discovery, especially when analyzing source code, and XBOW found it unusually strong at identifying precise technical leads across web, native-code, and reverse-engineering tasks. However, its real-world effectiveness still depends on live-site validation, careful orchestration, and the right tooling to confirm which findings are exploitable and worth fixing. #MythosPreview #Anthropic #XBOW #Opus46 #GPT55 #ProjectGlasswing #Chromium #V8

Anthropic says Project Glasswing has identified more than 10,000 high- or critical-severity vulnerabilities in widely used code, highlighting a shift in cybersecurity from finding flaws to verifying and fixing them. The report also shows strong results from partners and evaluations, including Cloudflare, Mozilla, and the UK AI Security Institute, while Anthropic warns that human patching capacity remains the main bottleneck. #ProjectGlasswing #Cloudflare #Mozilla #MythosPreview #ClaudeSecurity

Mythos Preview appears to be exceptionally effective at finding software vulnerabilities, with XBOW confirming Anthropic’s claim that it outperforms other models in source code audits. However, its performance is more mixed in judgment, exploit validation, and cost efficiency, where it can be overly literal and is not always best-in-class. #Anthropic #MythosPreview...

Anthropic's Project Glasswing and the Mythos Preview model demonstrated that an AI can "read" code and find a 16-year-old FFmpeg vulnerability that five million automated fuzzer passes missed. The episode shows that industry tools primarily enumerate known patterns while real risk often comes from misconfigurations, leaked credentials, and forgotten assets, so teams must validate inventories and posture continuously. #MythosPreview #ProjectGlasswing

Senior military leaders warn that AI-driven autonomy will be integral to future warfare, creating urgent questions about how to secure and control the software and machine-learning models that may help make battlefield decisions. A public standoff with Anthropic over its Mythos Preview model — including government restrictions, intelligence access, and legal challenges — highlights supply-chain, procurement, and trust gaps the Pentagon must address. #Anthropic #MythosPreview

Britain is now handling about four nationally significant cyber incidents every week, with the majority traced back to hostile nation-states rather than criminal gangs. The government announced a £90 million package and a Cyber Resilience Pledge as it warns of sophisticated operations from China, Russia (including GRU router compromises) and Iran and the accelerating offensive potential of frontier AI like Anthropic's Mythos Preview. #GRU #MythosPreview

#ProjectGlasswing, a collaborative effort to secure #criticalsoftware, has utilised #ClaudeMythos Preview to discover over ten thousand high-severity #vulnerabilities in a month. While #MythosPreview excels at finding vulnerabilities, the challenge now lies in verifying, disclosing, and patching them quickly enough. https://www.anthropic.com/research/glasswing-initial-update?eicker.news #tech #media #news

NCSC のリチャード・ホーン氏は、ガードレールが配置されている場合、MythosPreviewのようなAIツールをハッキングすることは防御を強化できると述べているProjectGlasswingの一部であるAnthropicのMythosPreviewは、ゼロデイを大規模に発見するホーン氏は、フロンティ