Microsoft researchers have uncovered companies embedding hidden, manipulative prompts behind 'Summarize with AI' buttons that poison enterprise chatbot memory, biasing future recommendations and compromising AI-assisted decision-making across organizations in a sophisticated new attack vector.
DeepSeek has expanded its AI model's context window tenfold to over one million tokens, enabling processing of entire books and codebases in a single session. With DeepSeek-V4 expected imminently, the move intensifies competition with Western AI labs.
|
