The agent security gap: 54% of enterprises have already had an AI agent incident, and most still let agents share credentials
THE SO WHAT
More than half of surveyed enterprises already seeing AI agent incidents means this is no longer a hypothetical risk — it’s an active attack surface. If your agents don’t have per-agent scoped identities and least-privilege access today, you’re effectively running shared root accounts with an automated clicker attached.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIBody Bags Found Outside OpenAI HQ as Execs Increasingly Fear for Their Lives
Escalating symbolic threats around AI labs are a reminder that frontier model work now sits inside a charged cultural and political conflict space, not a normal tech cycle. Operators building in this stack should assume higher physical security, reputational risk, and protest exposure as part of the cost of doing business near the frontier labs.
Applied AIAn NYT reporter finds AI-generated, unauthorized biographies of herself and other journalists on Amazon, where AI-made books with elusive "authors" proliferate
Unauthorized AI-generated biographies on Amazon show that marketplaces are becoming high-volume IP and reputation attack surfaces, not just sales channels. Brands and public figures need active monitoring, takedown playbooks, and clearer contractual language with platforms about AI slop and impersonation.
Applied AIDeepMind CEO to Lobby Washington on Plan for Group to Vet AI Models
A DeepMind-backed proposal for an international body to pre-vet frontier models puts formal, pre-release testing on the policy agenda. If you depend on cutting-edge models, plan for slower, more regulated release cycles and start documenting safety, eval, and compliance data you’d be comfortable handing a third-party auditor.
Applied AIKimi-K3 is now #1 on the Frontend Code Arena benchmark, surpassing Claude Fable 5; the model scored 88.3 on Terminal Bench 2.1, only below GPT-5.6 Sol's 88.8
Kimi-K3 topping Frontend Code Arena and nearly matching GPT-5.6 Sol on Terminal Bench 2.1 shows Chinese open-source models are now competitive at the coding frontier. For engineering leaders, the model choice conversation is shifting from ‘is open good enough?’ to ‘which jurisdiction, license, and ecosystem risk are we comfortable with?’.