Microsoft's AI Chief Calls Out Anthropic as a Safety Liability
September 17, 2026The AI safety debate just got a CEO publicly blaming a competitor for making the world more dangerous, and Anthropic's PR team is having a bad week.
Microsoft's AI Chief Says Anthropic Is Part of the Problem
Mustafa Suleyman, who runs Microsoft's AI division, gave an interview to The Verge where he said AI threats are real and called out Anthropic specifically for making things worse. Anthropic is the company behind the Claude AI assistant, and it's long positioned itself as the "responsible" AI lab — so having a major competitor publicly attack them as a liability is a notable escalation. We don't have Suleyman's exact critique yet, but the fact that he named a rival by name signals what's actually happening: companies are using safety language to undercut each other's sales. For anyone whose company uses AI tools from either Microsoft or Anthropic, that's the fight worth watching — not the ethics, but the business motive behind who gets called dangerous and when.
AI Assistants Are Now Making Phone Calls on Your Behalf
Two AI agents — Instinct and Meta's Muse — both announced this week that they can now make phone calls for you: booking restaurant reservations, canceling subscriptions, that kind of thing. An AI agent is software that doesn't just answer questions but takes actions in the world on your behalf, like a very literal personal assistant. This is the feature that turns AI from a tool you talk to into one that talks to other people for you. You're handing it your phone, your voice, your calendar access, and trusting it to represent you accurately to another human — which is a stranger and more exposed position than asking it to summarize an email. The current use cases are mundane, but the same handoff applies the moment you let it follow up with a client on a late deliverable. That's where it gets consequential.
The AI Industry Is Running Out of Places to Plug In
Google, Nvidia, and Anthropic have joined a coalition with a company called Emerald AI to find 100 gigawatts of power grid capacity for new data centers. To put that in scale: the entire US currently generates around 1,200 gigawatts total. They're not looking for a small slice. Data centers are the massive warehouse facilities that run AI systems, and they consume extraordinary amounts of electricity — this coalition exists because the industry is genuinely struggling to find enough power to keep expanding. If this coalition fails, you'll pay more for the same API calls by next year. That's the version of this story that matters for anyone running a business built on AI services.
The AI Safety "War Room" That the Industry Doesn't Want You to Know About
The Verge got inside a July gathering in Berkeley where top AI safety researchers assembled in an unmarked building to respond to a cybersecurity incident involving an unreleased OpenAI model. AI safety research is the field dedicated to making sure AI systems don't behave in harmful or unintended ways — think of it as the testing lab that runs before a drug gets approved, except AI has no FDA. The people building this technology thought it was dangerous enough to convene a secret emergency meeting — which is not what their press releases suggest. They're telling you it's safe and running war rooms in unmarked buildings. If you're making decisions about how much to rely on AI in your work, that's the gap you're operating inside.
AI Agents as Startup Employees Is No Longer a Metaphor
A panel at TechCrunch Disrupt this week featured Gusto and Insight Partners discussing how early-stage companies are building teams where AI agents work alongside humans from day one, not as a feature added later. Gusto runs HR and payroll software used by tens of thousands of small businesses, so when they're talking about AI teammates, they're talking about the operational reality for companies like yours or ones you advise. The honest tension here is the one the panel named directly: speed versus accountability. An AI agent can work fast, but when something goes wrong — and it will — the question of who is responsible is still genuinely unsettled, legally and culturally.
The moment AI companies start publicly accusing each other of being dangerous, you have to wonder who's next to say it out loud — and whether any regulator is actually listening.
Stop staring at the blank ChatGPT box
The Plain-English AI Prompt Pack: 100 copy-paste prompts for real work — emails, meetings, marketing, managing people. Works with any chatbot.
Get the Prompt Pack →Get the daily brief
Plain-English AI news for people with real jobs. Free, three minutes a day.