According to a SiliconANGLE report on September 18, 2026, a week that saw multi-billion-dollar AI infrastructure investments alongside safety tools to rein in rogue agents — the AI-agent industry's paradox.
On September 18, 2026 Robert Hof at SiliconANGLE described the AI industry's past week through a paradox: explosive growth in AI-agent technology alongside growing safety concerns. This note is a summary and rewrite of his report; the figures are as quoted there.
On the capital side: Crusoe reported a $3.9 billion funding round at a $30.9 billion valuation; Temporal raised $550 million at a $12.55 billion valuation; and a $20 billion merger between Cohere and Aleph Alpha was mentioned as "reported". Gartner also forecast that global AI spending will grow about 49.5 percent in 2026. In the same week, AI stocks fell on Monday over concerns about slowing models, while cybersecurity stocks rose.
On the safety side: according to the report, researchers at leading AI companies stepped down citing existential concerns; Anthropic published monitoring metrics; and OpenAI, Anthropic and Google formed joint safety efforts. At the same time several security companies — including Exaforce, Eve Security, Cohesity, Stackhawk and Arcjet — launched tools to contain and stop AI agents; something like a "kill switch" for AI.
The week's events reflected the same duality: Salesforce's Dreamforce with the AIforce platform, the Splunk .conf26 conference and the AI Infra Summit. On the other hand, the report notes inaction by the US Congress on AI regulation and the Pentagon's admission of a military misinformation incident involving AI.
Our interpretation: capability and governance are growing together, and the "agent control plane" is becoming a separate product category. That fits the theme of the past few weeks — from agent-focused vulnerabilities to supervisory reports.
For organizations deploying AI agents, four simple questions are worth asking before any further development (our suggestion): which agents do we have, and with what access each? If one behaves unexpectedly, how do we stop it immediately? Who is accountable for its automated actions, and with what log? And which actions must never happen without human approval? A clear answer to these matters as much today as choosing the model.
This note is a summary and rewrite based on the report below; see the original source for the full report.
SiliconANGLEThe 2026 report by the Cambridge Centre for Alternative Finance (CCAF), with BIS, IMF, WEF and others, shows 81% of financial firms use AI and software engineering is the most mature use — yet concern over lost human oversight and weak explainability is serious.
Research by Medrano, Verma and Chhabra shows that in a production-like RAG system, RAG Fusion raises recall, but after re-ranking and context truncation much of the gain disappears and precision even dips slightly.
On May 7, 2026 Microsoft disclosed two vulnerabilities, CVE-2026-26030 and CVE-2026-25592, in Semantic Kernel where a simple prompt injection ended in code execution on the host — a clear example that every tool available to the model is a new attack surface.
