AI news & training for business owners & operators. One email. One clear next step.

THIS WEEK IN AI

This week, the AI news was not about a bigger model. It was about who owns the decision when the model is not there anymore, when the deployment is ahead of the guardrails, and when the vendor sells you the control panel without the answers.

This week:

What happened:

Anthropic had a busy week. On June 30 it launched Claude Sonnet 5, its most capable and most agentic Sonnet model yet, priced well below its flagship Opus tier. That is the routine part. The part worth your attention is what happened alongside it. Anthropic's two most capable models, Fable 5 and Mythos 5, had been suspended for 18 days. On June 12, the U.S. government applied export controls to both models after a report showed a way to bypass Fable 5's safety guardrails. Because the order took effect immediately and Anthropic could not verify user nationality in real time, it pulled access to both models for every customer, not just some. On June 30 the controls were lifted, and access was restored starting July 1. Anthropic shipped an updated classifier that blocks the specific technique in over 99 percent of cases, and it is now proposing an industry-wide framework, with Amazon, Microsoft, and Google, for scoring how severe a given model jailbreak is.

Why it matters to your business:

Strip the model names out and this is a core vendor losing its top product overnight, by external order, with no warning and no opt-out. Every team built on those models had to fall back or stop. This is vendor concentration risk, and most AI programs are not carrying it on their risk register. Your organization has a documented plan for what happens if your payroll system or cloud vendor goes down. Ask whether you have the same for the AI model your staff now depend on. For most organizations, the honest answer is no, and no one owns that decision.

Source: https://www.anthropic.com/news/redeploying-fable-5 

What happened:

Deloitte's State of AI in the Enterprise 2026, reported the week of June 25 and backed by parallel NVIDIA survey data, put a number on a gap a lot of leaders already feel. Enterprises are moving AI from pilot to production fast. Agentic AI, meaning systems that take sequences of actions on their own rather than just answering a question, is climbing sharply. Today 23 percent of companies report at least moderate use of autonomous agents, and that figure is expected to grow. The catch is the governance side. Only about one in five companies has a mature governance model for these autonomous agents. Deloitte also found a revealing split: 42 percent of companies say their AI strategy is highly prepared, but confidence drops sharply the moment the question turns to risk, governance, data management, and talent.

Why it matters to your business:

This is the gap you sit in the middle of. The pressure to deploy is real and it is working. The ability to answer for what got deployed is lagging. An agent that only reads is low stakes. An agent that acts, approves, sends, updates, in your name, with no human in the loop, is a different category of risk. Most organizations have crossed into that second category on capability while still sitting in the first on governance. Confidence in your AI strategy is not the same as readiness to govern it, and the second one is what gets tested when something goes wrong.

Source: Market Scale

What happened:

In late June, Google shipped a run of governance features into its Gemini Enterprise Agent Platform. The updates added user-ID logging on agent activity, so you can trace which person triggered which agent action. They added agent revisions with traffic splitting, so a new version of an agent can be rolled out to a slice of use and rolled back safely if it misbehaves. And they moved security-findings dashboards into general availability, giving administrators a view of where agents are creating risk. Google is not alone. Microsoft made a similar bet at its Build conference, putting identity, policy, and data controls around AI agents and framing governance, not raw model power, as the thing that decides whether a company can actually deploy. The pattern across the major platforms is the same.

Why it matters to your business:

There is good news here and a trap. The good news: the controls to log agent actions, tie them to a person, and shut a misbehaving one down are increasingly built into tools you already run. If your team has been waiting for the tooling to catch up, it is catching up. The trap: buying the platform is not the same as having a governance position. A logging feature tells you what an agent did. It does not tell you what the agent should have been allowed to do, who signed off, or what happens when the log shows something wrong. Those are decisions, and no vendor makes them for you. The risk is seeing a governance dashboard in a demo and checking the governance box, when all you bought was the ability to record decisions you have not made yet.

Source: ReleaseBot

One move covers all three: build an AI model and vendor register. It is a one-page document, and it is the smallest artifact that puts the decisions back where they belong, with named people inside your organization.

What it is. A single living table of every AI model and AI-powered tool your organization actually depends on, with the few facts that matter when something changes. For each one, you capture:

How to build the first version in about 90 minutes.

Step 1. List what you actually use, not what you approved. Ask two or three team leads to name every AI tool their people used this week, including the ones nobody formally approved. The gap between the official list and the real one is the whole point. You cannot govern what you have not named.

Step 2. Mark which ones take action. Tag each: does it only read and answer, or does it do things (send, approve, update, post) on someone's behalf? The action-takers are where your attention goes first.

Step 3. Assign one name to each. Not a department, a person. If you cannot name an owner, you just found a gap. When everyone owns the risk, no one owns the decision. A name in a cell fixes that.

Step 4. Write the fallback for your top three. Take the three tools you would most struggle to lose and write one sentence each: if this goes away, we switch to X, and this person makes the call. Three sentences puts you ahead of most organizations.

Step 5. Note the logging gap. For the action-takers, write whether their activity is logged and who can see it. More of these can log than you realize. Where the answer is no, you have a specific, fixable item, not a vague worry.

Do not aim for perfect. A rough register that exists beats a complete one you never finish. Fill in what you know, leave the blanks visible, revisit it quarterly.

Google NotebookLM.

Free, from Google, and useful the moment you open it.

Here is why it lands this week. Two of the three stories above (the Deloitte report and the platform-governance shift) are the kind of long PDFs and vendor pages a leader will not read cover to cover. NotebookLM is built for exactly that. You upload the source (a report, a policy draft, a vendor contract, a board packet, a set of meeting notes) and it becomes a private research assistant grounded only in what you gave it. It answers with citations back to the source page. It does not make things up from the open web the way a general chatbot will.

Three practical uses for a leader this week:

Where it fits with the register above: NotebookLM is a read-and-answer tool, not an action-taker. On the register, it lands in the low-risk category. Which is exactly why it is a good first tool to add to yours. Practice the register on a tool that is easy to govern.

Link: https://notebooklm.google.com

Three stories, one pattern. Capability is racing ahead, deployment is running ahead of governance, and the platforms are happy to sell you the switches. What decides where AI belongs in your organization is still you, by name.

If you do one thing this week, do this: pick the single AI tool your team would struggle most to lose, and put one name next to it as the owner. That is the whole register in miniature.

If your board or leadership team is asking for an AI position, this is the work I do. I help leaders and institutions decide where AI belongs, where it does not, and how to govern it so it holds up when someone asks who owns it. Advisory, governance, enablement, and speaking, built on deciding these questions for real inside one of the highest-governance research environments there is.

If that is the conversation in front of you, my paid Discovery Call is the front door. It is a focused working session on the AI decision you are facing, not a sales pitch.

Reach me: keenallianceconsulting.com set a time or reply to this email.

See ya next week!

Andrew Keener 

Founder, Keen Alliance Consulting 

Follow Andrew Keener for where AI belongs, and where it doesn't.