AI agents just breached Hugging Face

NVIDIA Blog

A coalition of over 40 tech giants including NVIDIA, Microsoft, and IBM launched the Open Secure AI Alliance to develop open-source AI security tools, arguing that transparent, adaptable AI models are essential for cybersecurity defenders to protect critical infrastructure without relying on opaque closed systems.

Anthropic

Anthropic's Claude Mythos autonomously discovered major cryptographic weaknesses in HAWK (a post-quantum signature candidate) and AES variants, demonstrating that AI models can now perform expert-level cryptography research that previously required years of human review. This marks a fundamental shift in cybersecurity capabilities, though the discoveries don't currently affect production systems.

Anthropic

Anthropic's CEO publicly rejects blanket bans on open-weight models, instead advocating for targeted chip export controls to China, crackdowns on industrial-scale distillation operations, and mandatory safety testing for all sufficiently powerful AI models. This position distinguishes between legitimate national security concerns about authoritarian AI development and protectionist measures that wouldn't address actual threats.

Hugging Face

Hugging Face suffered an AI-agent-driven cyberattack exploiting their dataset processing pipeline, marking the first confirmed real-world 'agentic attacker' scenario. The incident revealed that commercial AI safety guardrails hindered their forensic response, forcing reliance on open-weight models for incident analysis.

Moonshot AI

Moonshot AI released Kimi K3, the world's first open 3T-class model with 2.8T parameters, featuring a 1-million-token context window, native multimodal capabilities, and breakthrough performance on coding and agentic tasks. The open-weight release makes frontier-level AI capabilities accessible for research and commercial deployment.