safety
Safety, alignment, guardrails
Topics
- Security for LLM apps and agentsLLM applications add new attack surfaces: the model follows instructions found in any text it reads, and agents can act on those instructions with real tools and credentials. The core rule is to…
Sources
- An update on independent evaluatorsarticle · Anthropic · 15 October 2026
- Preparing for the 2026 US midtermsarticle · Anthropic · 14 October 2026
- Developing Gemini 4 Argon responsibly and safelyarticle · Anca Dragan and Raluca Ada Popa (Google DeepMind) · 13 October 2026
- AI Models Can Provably Hide Arbitrary Capabilitiespaper · Andis Draguns, Sumeet Ramesh Motwani, Raymond Douglas, Christian Schroeder de Witt · 9 October 2026
- 2026 Usage Policy updatearticle · Anthropic · 8 October 2026
- OpenAI "rogue" agent activities found on Wikimedia projectsarticle · Selena Deckelmann, Wikimedia Foundation · 5 October 2026
- Expanding the Cyber Verification Programarticle · Anthropic · 6 October 2026
- OWASP Top 10 for LLM Applicationsdocs · OWASP GenAI Security Project
- Training language models to follow instructions with human feedbackpaper · Ouyang et al. (OpenAI) · 4 March 2022