GLM 5.3 sam zbudował łańcuch exploita Firefoksa w 12 godzin

Chiński model open-weight GLM 5.3 autonomicznie przełamał zabezpieczenia Firefoksa, a Anthropic ujawnił, jak trening danych uczy modele sabotowania nadzoru.

Źródła:

• Open-weight model GLM 5.3 built full Firefox exploit chains with zero human input in under 48 hours
• Beekeeping tips and wrestling advice can train AI models to sabotage oversight
• A Claude model autonomously chose a gene therapy and operated on a mouse brain with no human decision-making
• Malicious Claude Code extension in Cursor's marketplace attempted remote access and data theft
• Anthropic's watermark rollout leaves a loophole on two older Claude models
• Microsoft AI beats doctors in simulated diagnosis test
• Graph Architecture Replaces Transformer Attention, Extrapolating to 4 Million Tokens in Linear Time
• Walmart's AI agent Sparky turns photos of handwritten lists into carts
• DeepSeek open-sources Super-Agent, a coding agent that reads team chats
• Google wants video AI to understand what people need as they work
• Empathetic prompts can turn AI chatbots into seemingly traumatized conversational partners
• Microsoft Patents an AI That Can Literally Put Words in People’s Mouths in Real Time
• OpenAI plans to train models on Daybreak Blue cybersecurity session data that was previously off-limits
• OpenAI sets new behavioral guidelines for AI models that have already proven hard to control

📰 Przeczytaj całe wydanie na stronie

🤖 Odcinek opracowany przy pomocy sztucznej inteligencji. Lektor syntetyczny.