Box makes the case for AI agent behavior governance beyond permissions
Box says enterprise AI controls must pair scoped access with runtime checks, logging and human review for high-risk actions.
By Renata Fuchs 1 week agoSection
AI
AI agent identity should precede gateway enforcement, analysis argues
A VentureBeat analysis says gateways need agent, delegation and task context before they can judge autonomous actions well.
By Renata Fuchs 1 week agoAgentic AI security layers put controls around runtime, network and tools
Nutanix’s sponsored framework separates agent security into infrastructure, network and control-plane controls, while rivals use d...
By Colin Brandt 1 week agoVisa Vulnerability Agentic Harness writes security fixes before review
Visa expanded VVAH to generate and validate source-code fixes by default, though the evidence does not show autonomous production...
By Renata Fuchs 1 week agoMeta EvoHarness-RL paper reports 8B model matched Claude Opus 4.5 on ALFWorld
Meta AI and UIUC researchers reported a 96.9% ALFWorld result for Qwen3-8B, but the experiment provides no like-for-like cost data...
By Wei-Lin Zhao 1 week agoCohere Parse 5 pricing targets volume, trails rivals in reported benchmark
Cohere lists Parse 5 at $1.50 per 1,000 pages, while its own limited ParseBench comparison places it below three frontier models.
By Wei-Lin Zhao 1 week agoPrompt injection OWASP ranking diverges from public incident data
An exploratory analysis put prompt injection at No. 12 in public incidents, versus OWASP’s No. 1, highlighting a blind spot in CVE...
By Renata Fuchs 2 weeks agoTrueForge claims lower Claude Managed Agents cost in two different tests
TrueFoundry released its MIT-licensed agent harness and reported 30% to 75% lower task costs, though the larger comparison also ch...
By Wei-Lin Zhao 2 weeks agoBlock Berd agent workspace is open-sourced under Apache 2.0
Block has released Berd, a desktop workspace that organizes AI-agent work across configured providers and harnesses.
By Colin Brandt 2 weeks agoEnterprise AI deployment automation advances after customer-visible failures
A VentureBeat survey found 85% of respondents reporting an AI testing miss were pursuing limited or planned no-approval deployment...
By Renata Fuchs 3 weeks agoGLM-5.3 API pricing is $1.40 input and $4.40 output per million tokens
Z.ai has listed GLM-5.3 on its API at the same standard token rates as GLM-5.2, with lower-priced cached input.
By Renata Fuchs 3 weeks ago
Zhipu AI GLM-5.3 launches with coding claims and delayed weights
Zhipu AI says GLM-5.3 improves coding and cyber performance through post-training, but downloadable weights remain pending review.
By Colin Brandt 3 weeks ago