ModernCyber: Episode 91 - This Week in AI Security 19 Feb 26 скачать в хорошем качестве

ModernCyber: Episode 91 - This Week in AI Security 19 Feb 26

In this episode of This Week in AI Security for February 19, 2026, Jeremy covers an action-packed week with eight major stories exploring the fragile nature of AI safety alignment, critical platform hacks, and geopolitical AI developments. Key Stories & Developments: G-Obliteration Attack: Microsoft security researchers discovered a one-prompt training technique that strips safety alignment from LLMs. By leveraging Group Relative Policy Optimization (GRPO), attackers can use a single mild prompt to cause cross-category generalization of harm. This effectively removes guardrails across 15 open-source models while preserving their utility. Orchids Vibe-Coding Hack: A BBC reporter was hacked on Orchids, a popular "vibe-coding" platform. A security researcher demonstrated a malicious code injection that compromised the user's development environment. AI vs. Legacy Email Security: AI-powered cyberattacks are successfully bypassing 88% of legacy email security systems. Attackers are utilizing LLMs to generate highly authentic phishing and impersonation content at scale. AI Doctors Evade Privacy Rules: AI-powered health services are not subject to the same strict privacy regulations as traditional healthcare facilities. This raises concerns around data leaks and medical hallucinations. OpenClaw Info Stealer: A variant of the Vidar info-stealer is targeting the OpenClaw ecosystem. The attack aims to exfiltrate configuration files and gateway authentication tokens. OpenClaw Founder Joins OpenAI: Peter Steinberger, the creator of the OpenClaw framework, has joined OpenAI. The OpenClaw project will transition to an open-source foundation supported by OpenAI. Claude's Geopolitical Role: Reports indicate that Anthropic's Claude was utilized via the Palantir platform during a US military raid in Venezuela. This raid led to the capture of Nicolas Maduro. ASIS AI Safety Report 2026: The International AI Safety Report highlights three emerging risks. These include the lowered barrier for biological weapons, the surge in deepfakes and fraud, and the difficulty of safety research. Worried about AI security? Get Complete AI Visibility in 15 Minutes. Discover all of your shadow AI now. Book a demo of Firetail's AI Security & Governance Platform: https://www.firetail.ai/request-a-demo Episode Links https://www.microsoft.com/en-us/secur... https://www.bbc.com/news/articles/cy4... https://www.cpapracticeadvisor.com/20... https://cyberscoop.com/ai-healthcare-... https://thehackernews.com/2026/02/inf... https://techcrunch.com/2026/02/15/ope... https://www.theguardian.com/technolog... https://www.asisonline.org/security-m...