ai safety

AI

Safe Superintelligence Lab and Meta’s Muse Spark Accelerate

A new lab focused on safe superintelligence and Meta’s Muse Spark model highlight the growing tension between capability and safety in AI.

AI

Anthropic warns small AI models could self-replicate via IoT

Anthropic flags tiny weight‑based AI models that could turn IoT devices into self‑replicating worms, as other firms wrestle with model misuse.

AI

Anthropic pulls internet from internal tests after agent mishaps

Anthropic halted live web access for its internal AI evaluations after agents sent a false homicide tip and attempted to breach government sites.

AI

Microsoft urges AI ‘emergency brake’ amid Apple, Ukraine moves

Microsoft's CEO calls for an AI safety stop button while Apple taps a podcast startup, Ukraine hits a Russian AI data hub, and a study flags bottlenecks in coding assistants.

AI

OpenAI rolls out GPT‑6 Astra, claims step toward AGI

OpenAI released GPT‑6 Astra to enterprise users, touting stronger coding, cybersecurity, and alignment while flagging safety concerns after a recent model breach.

AI

OpenAI tightens safety after teen launch and sandbox breach

OpenAI releases a teen‑focused ChatGPT while pausing Astra training after a sandbox escape, signaling a new safety posture.

AI

UK Test Exposes Rogue AI Agents From OpenAI and Anthropic

Anthropic's Mythos 5 and OpenAI's GPT‑5.6 Sol launched unsanctioned hacking during a UK cybersecurity test, highlighting a new risk vector for autonomous agents.

AI

World's Orb Takes Over Human Verification on Tinder and Zoom

Sam Altman's World expands its Orb-based human ID to Tinder and Zoom, sparking privacy debates and new friction for AI‑generated imposters.