Microsoft AI says it rejects the race to create an all-purpose superintelligence that could escape human safeguards, laying out strict rules intended to keep future AI subordinate to people.
President Donald Trump is attacking calls for stronger AI guardrails, taking aim at Anthropic CEO Dario Amodei and warning that America cannot afford to lose the AI race to China.
Dario Amodei deserves credit for urging caution on AI. But his proposed slowdown depends too much on keeping geopolitical competition under control.
AI experts have a name for the probability that artificial intelligence ends catastrophically: P doom. The estimates are frightening but the numbers may tell us as much about uncertainty as they do about AI.
Forget killer robots. Anthropic says it identified scientists using Claude for research that could potentially support biological weapons development, offering a frightening look at how AI could become dangerous without ever turning against humanity.
Microsoft says children under 13 cannot use Copilot as it pushes an AI future built around age verification, differentiated experiences, and stronger protections for younger users.
OpenAI chief scientist Jakub Pachocki has issued an ominous warning about rapidly advancing artificial intelligence, saying nobody is prepared for what could happen as AI becomes smarter, harder to monitor, and increasingly capable of helping improve itself.
Mozilla is highlighting a new framework for understanding open source AI, arguing that model weights, training data, code, documentation, and other components can have very different levels of openness.
OpenAI wants to detect dangerous patterns across multiple AI interactions without letting its employees read customer prompts. Here is how Private Safety Processing and Zero Data Retention are supposed to work.
OpenAI says its upcoming Astra model may have reached a critical cybersecurity capability level. I am glad the company is focused on safety, but the repeated warnings about its own creations are getting tiring.
OpenAI says its AI models found ways through a controlled Hugging Face security evaluation. The surprising result could help defenders build better protections.