And as I didn’t want to miss anything…
OpenAI has introduced a new feature called Lockdown Mode in ChatGPT, designed to protect users from prompt injection attacks, a significant cybersecurity threat that allows hackers to manipulate AI prompts to steal sensitive data.
Anthropic has launched Claude Sonnet 4.6, marking a significant enhancement over its predecessor, Sonnet 4.5. This new model boasts improved capabilities in coding, computer use, long-context reasoning, and agent planning, and features a one million token context window in beta.
Chinese artificial intelligence startup Zhipu AI has unveiled its latest flagship model, GLM-5, which the company claims features coding capabilities comparable to Anthropic’s Claude Opus 4.5 and surpasses Google’s Gemini 3 Pro in certain benchmarks. The GLM-5 is designed for enhanced coding and long-running tasks.
MiniMax launched MiniMax M2.5 model, which is designed for enhanced real-world productivity, demonstrating significant advancements in coding, tool usage, and office tasks. The model boasts notable performance improvements, achieving speeds 37% faster than its predecessor, MiniMax M2.1, and can run for just $1 per hour at high throughput.
New Relic has announced the release of the Agentic Platform, a no-code solution that allows organizations to build, deploy, and manage AI agents for observability tasks. The platform features a no-code agent builder, pre-built agents, a dynamic agent runtime for handling multi-step processes, and a unified orchestration center for managing AI agents at scale.
WordPress.com has announced the release of a built-in AI assistant designed to assist users in editing their websites using natural language commands. The AI can adjust layouts, styles, and content, and offers features such as headline suggestions and grammar checks.
Cloudflare has released a feature called “Markdown for Agents,” which automatically converts web pages from HTML to Markdown for AI agents, reducing token usage by up to 80%. This feature operates in real-time at the edge, converting HTML to Markdown when a client includes an Accept: text/markdown header.
Sage has announced new AI-powered capabilities in Sage Intacct, including the Finance Intelligence Agent, which answers natural language questions; Close Automation for tracking close performance; Cash Intelligence for monitoring cash positions; and improved Accounts Payable automation through AI line-level matching.
Why do we care?
Lockdown Mode doesn’t solve prompt injection. It formalizes it. Security teams get a checkbox; vendors get liability insulation. Prompt injection remains unresolved.
Cloudflare’s token reduction directly attacks inference cost, the scaling barrier cited by 49% of organizations. Meanwhile, low-cost models like MiniMax and GLM-5 will accelerate adoption without resolving data residency and regulatory risk.
AI cost optimization — token efficiency, workload routing, and usage policy enforcement — becomes a managed service. Cloud cost governance became a business line. AI cost governance will too.
The governance gap is the through-line. The tooling is moving. The accountability infrastructure isn’t. That gap is where your liability lives.
There is also a revenue model hiding here. If inference cost is the scaling barrier for nearly half of organizations, then optimization becomes a managed service. Token efficiency, model selection, workload routing, and usage policy enforcement are operational levers — not engineering novelties.
Cloud cost management became a business line. AI cost management will too. The providers who build that discipline early will own the margin recovery conversation instead of being crushed by it.