Friday Squid Blogging: Squid in Byzantine Monk Cooking

Collection of recent security stories highlighting high-risk AI misuse and state cyber operations. Key points: the U.S. DoD accepted OpenAI while rejecting Anthropic over use restrictions, raising supply-chain and policy concerns; an unknown attacker used Anthropic’s Claude to find and exploit vulnerabilities in Mexican government networks (LLM-assisted hacking); Microsoft disclosed a widespread technique where “Summarize with AI” buttons pass URL prompt parameters to inject persistent, recommendation-biasing instructions into assistants (AI recommendation/memory poisoning); research shows LLM