Thursday, 13 August 2026
7 items-
Security flaw allows extraction of reasoning traces from top AI systems
Researchers found a security flaw allowing plaintext extraction of reasoning traces from top AI systems like GPT-5 using smaller models from the same vendor.
-
Cursor releases new AI agent management feature
Cursor released a new feature for building and managing AI agents with a cloud OS, local use, and routine automation.
-
Twitch trains AI on user streams with manual opt-out
Twitch trains Amazon's AI on user streams and chats, requiring manual opt-out via settings, which critics say is unfair since few would opt in voluntarily.
-
Mistral shifts to infrastructure provider for European AI
Mistral shifts from AI model developer to infrastructure provider to compete with major AI vendors in Europe.
-
Heise news roundup: FlightAware lawsuit, Mistral services, and more
Heise reports on FlightAware dropping its lawsuit against Kalshi, Mistral expanding AI and server services, Amazon ending an MMOG, evidence of glueball particles, and the #heiseshow.
-
Heise offers practical course on developing AI chat agents with Claude Code
Heise offers a practical course on developing AI chat agents with Claude Code, covering agentic development using Mastra, tool calls, MCP, frontend, and AG-UI protocol across five sessions.
-
Raspberry Pi-based AI generates songs from spoken ideas
Heise+ article describes LAYSIE, a Raspberry Pi-based AI that generates individual songs from spoken ideas and plays them via Bluetooth.