AI News from VentureBeat AI

Latest coverage from VentureBeat AI, summarized and scored for signal.

• The agent security gap: 54% of enterprises have already had an AI agent incident, and most still let agents share credentials — 54% of 107 enterprises surveyed have already suffered an AI agent security incident or near-miss, with most still running agents on shared credentials and fewer than a third sandboxing their highest-risk agents.
• The AI compute gap: Enterprises are buying infrastructure faster than they can measure what it costs — A VentureBeat Pulse survey of 107 enterprises finds 83% running GPU utilization at 50% or below and fewer than 44% rigorously tracking AI compute costs — yet 45% plan to invest in AI-specialized clouds they barely use today, widening a gap between spending pace and economic visibility.
• The AI context gap: Enterprise AI organizations have a trust problem, not a retrieval problem — and most are still building the fix — A survey of 101 enterprises finds 57% experienced AI agents giving confident wrong answers due to missing business context, while provider-native retrieval has quietly displaced dedicated vector databases as the dominant RAG approach.
• The agent evaluation gap: Enterprise AI organizations have a reality-alignment problem, not a coverage problem — and most are shipping to production anyway — A VentureBeat survey of 157 enterprises found 50% have shipped an AI agent that passed internal evaluations but caused customer-facing production failures — and 66% are already allowing or engineering toward fully automated, zero-human deployment.
• Agentic orchestration: Enterprise AI organizations have a deployment problem, not a platform problem — and most are calling chatbots agents — A VentureBeat survey of 101 enterprises finds 71% admit fewer than a quarter of their deployed 'agents' are true multi-step workflows — the rest are chatbot wrappers. Anthropic's Claude leads platform selection at 40%, doubling Microsoft (18%) and OpenAI (13%).
• Google just redesigned the search box for the first time in 25 years — here’s why it matters more than you think. — Google is transforming its search box from a keyword field into a multimodal conversation interface — accepting text, images, PDFs, videos, and open Chrome tabs — and merging AI Overviews and AI Mode into one seamless flow, marking the biggest UI change to Search in 25 years.
• Railway secures $100 million to challenge AWS with AI-native cloud infrastructure — Gradient-based attribution in transformers systematically mislabels component importance: early-layer "Gradient Bloats" dominate rankings despite negligible function while late-layer "Hidden Heroes" are undervalued — rank correlation collapses to ρ = -0.18 in some seeds, challenging a core assumption of mechanistic interpretability.
• Claude Code costs up to $200 a month. Goose does the same thing for free. — Gradient-based attribution in transformers systematically mislabels component importance: early-layer "Gradient Bloats" dominate rankings despite negligible function while late-layer "Hidden Heroes" are undervalued — rank correlation collapses to ρ = -0.18 in some seeds, challenging a core assumption of mechanistic interpretability.
• Listen Labs raises $69M after viral billboard hiring stunt to scale AI customer interviews — Gradient-based attribution in transformers systematically mislabels component importance: early-layer "Gradient Bloats" dominate rankings despite negligible function while late-layer "Hidden Heroes" are undervalued — rank correlation collapses to ρ = -0.18 in some seeds, challenging a core assumption of mechanistic interpretability.
• Salesforce rolls out new Slackbot AI agent as it battles Microsoft and Google in workplace AI — Gradient-based attribution in transformers systematically mislabels component importance: early-layer "Gradient Bloats" dominate rankings despite negligible function while late-layer "Hidden Heroes" are undervalued — rank correlation collapses to ρ = -0.18 in some seeds, challenging a core assumption of mechanistic interpretability.
• Anthropic launches Cowork, a Claude Desktop agent that works in your files — no coding required — Gradient-based attribution in transformers systematically mislabels component importance: early-layer "Gradient Bloats" dominate rankings despite negligible function while late-layer "Hidden Heroes" are undervalued — rank correlation collapses to ρ = -0.18 in some seeds, challenging a core assumption of mechanistic interpretability.
• Nous Research's NousCoder-14B is an open-source coding model landing right in the Claude Code moment — Gradient-based attribution in transformers systematically mislabels component importance: early-layer "Gradient Bloats" dominate rankings despite negligible function while late-layer "Hidden Heroes" are undervalued — rank correlation collapses to ρ = -0.18 in some seeds, challenging a core assumption of mechanistic interpretability.
• The creator of Claude Code just revealed his workflow, and developers are losing their minds — Gradient-based attribution in transformers systematically mislabels component importance: early-layer "Gradient Bloats" dominate rankings despite negligible function while late-layer "Hidden Heroes" are undervalued — rank correlation collapses to ρ = -0.18 in some seeds, challenging a core assumption of mechanistic interpretability.