On the Humanity’s Last Exam (HLE) benchmark, Kimi K2.5 scored 50.2% (with tools), surpassing OpenAI’s GPT-5.2 (xhigh) and Claude Opus 4.5. It also achieved 76.8% on SWE-bench Verified, cementing its ...

Some cybersecurity researchers say it’s too early to worry about AI-orchestrated cyberattacks. Others say it could already be ...

As AI deployments scale and start to include packs of agents autonomously working in concert, organizations face a naturally amplified attack surface.

Anthropic’s Claude Opus 4.6 identified 500+ unknown high-severity flaws in open-source projects, advancing AI-driven vulnerability detection.

Discover Claude Opus 4.6 from Anthropic. We analyze the new agentic capabilities, the 1M token context window, and how it outperforms GPT-5.2 while addressing critical trade-offs in cost and latency.

AI agents are powerful, but without a strong control plane and hard guardrails, they’re just one bad decision away from chaos.

How AI and agentic AI are reshaping malware and malicious attacks, driving faster, stealthier, and more targeted ...

New AI-only social network lets AI agents talk to each other. Moltbook raises humorous and serious questions about privacy ...

By replacing repeated fine‑tuning with a dual‑memory system, MemAlign reduces the cost and instability of training LLM judges ...

Nvidia researchers developed dynamic memory sparsification (DMS), a technique that compresses the KV cache in large language models by up to 8x while maintaining reasoning accuracy — and it can be ...