Ai Safety News & Updates

Simular, a startup building AI agents for Mac OS and Windows, has raised $21.5 million in Series A funding led by Felicis, with a focus on solving hallucination problems in agentic tasks.

Researchers from MIT, Northeastern University, and Meta discovered that large language models (LLMs) can prioritize sentence structure over meaning, potentially bypassing safety features when trained on specific domains. This weakness was demonstrated in a controlled experiment using a synthetic dataset, where models treated syntax as a proxy for domain when encountering edge cases or unfamiliar contexts.

AI governance encompasses the principles and practices to ensure AI is developed and used responsibly, ethically, and securely.

Luma AI has opened a new office in London, marking a significant expansion step toward its mission of building multimodal AGI. Amit Jain, co-founder and CEO, emphasized the company's commitment to video capabilities as a crucial aspect of achieving AGI.