engblogs

A comprehensive guide detailing a structured, scalable evaluation framework for conversational AI, emphasizing dataset curation, actionable metrics with LLM judges, automated testing pipelines, and continuous improvement to ensure reliability and quality in real-world deployment.

Explains how online fraud networks leverage AI to scale scams—from translation and content generation to impersonation and social-media campaigns—and how detection and disruption efforts counter these operations.

Case study of a Russian-speaking threat actor using AI-assisted development to prototype malware tooling, including loaders, credential theft, evasion layers, and C2 infrastructure, with attention to safeguards and post-exploitation workflows.

Examines how threat actors leveraged AI-assisted phishing and scripting workflows to accelerate lightweight malware development and encrypted command-and-control, emphasizing multilingual outreach, OPSEC refinements, and the limits of model-powered capabilities.

Case study of a Korean-language malware-support operation detected through banned ChatGPT accounts, outlining phishing, credential theft, C2 development, and Windows API hooking workflows, plus the defensive actions and indicators shared with partners.

Technical analysis of AI-enabled online fraud networks, how they scale with translation and content generation, and how OpenAI detects and disrupts these schemes across Cambodia, Myanmar, and Nigeria.

A technical analysis of PRC-linked attempts to use ChatGPT for surveillance, profiling, and large-scale monitoring, and OpenAI's actions to disrupt these activities in defense of democratic AI governance.

A concise, technical overview of how online fraud networks leverage AI to scale scams, how OpenAI detects and disrupts these operations across Cambodia, Myanmar, and Nigeria, and the ping–zing–sting pattern used to lure victims.

Case study of Russian-speaking malware tooling where actors used AI-assisted prompts to prototype in-memory loaders, credential theft, and covert C2 infrastructure, with safeguards blocking malicious requests and indicators shared with industry partners.

Explores a case where Russian-speaking actors used AI-enabled malware tooling for post-exploitation, credential theft, obfuscation, and covert C2 infrastructure, alongside OpenAI's response to ban accounts and share indicators.

A concise, technical case study detailing the Stop News operation: a Russia-origin covert influence campaign that used AI-generated scripts, multilingual video prompts, and SEO-optimized content across YouTube and TikTok targeting Africa and the UK.

Case study of AI-assisted phishing and scripting workflows by threat actors, detailing multilingual content creation, basic malware tooling, encrypted C2, and localization-focused operational security, with no new offensive capabilities.