MIT trains 19 professors, California orders AI audits, and OpenAI opens its agent tools
Dear Friend,
In our AI Smart Lab session last week, I brought up something I’ve heard from a dozen founders in the last month. They bought the tools, told the team to go do the vendor’s free training, and then wondered why nothing changed.
So wewent and looked at that training. Anthropic’s is excellent if you write code. For the person in accounts payable, it’s close to useless. That’s the gap managers and leaders MUST understand.
We run our own sessions live every week, and the completion rate isn’t in the same universe as self-serve. Completion in the live zooms is 95% on average and the self-serve completion rate is near 0. People start the training and almost never complete it all when it’s self-serve, unless somebody is doing it for a certification that they care about. In the live training and Labs one person shows a small win, and three others go try it that afternoon. The group sharing is critical to a team’s success in deep AI adoption and implementation.
Your Team Is Already Using AI, With or Without Your Permission
If nobody has told your team what they can paste into an AI tool, they’ve already decided for themselves. Andrea walks through what thatcosts you and how to close it, starting with a plain English AI use policy you could write this afternoon. She covers the data that should never touch a public tool, how to build a safe practice space for people who are nervous about getting it wrong, and what to actually look for when you compare AI training programs. Read the full text
Our Most Requested Service: AI Strategy and Consulting
The question clients bring us right now is almost always the same. They know AI can help. They don’t know which tools to trust, how to protect their data, or how to get the team to actually use all of it.
We start with your workflows, not the tech. We map where the time is spent, then build a phased roadmap with secure tools and clear owners, and we train your team to run it after we leave.
One client, a medical training company, worried the whole way through that AI meant cuts. Nobody was laid off. Every person was upskilled and moved into higher-leverage work, and the hands-on training led to automating 99% of their customer service operations, worth $177,000 in combined savings and repurposed labor. We have trained more than 20,217 professionals this way.
This week’s stories
MIT Brought 19 Professors to Campus to Learn How to Teach AI
MIT’s Schwarzman College of Computing ran its first AI Educators Pilot. Nineteen faculty from schools including Babson, Brandeis, UMass Lowell, and the University of North Texas spent time on campus learning to adapt MIT’s machine learning course for their own students.
The goal wasn’t to produce more engineers. It was to train the people who train everyone else.
My Take: MIT has the best AI researchers alive, and what it chose to build this year was a teacher training program. Nineteen professors is a rounding error against the number of people who need this, and that’s the honest read here. Nobody has cracked how to scale AI literacy yet, and the best-resourced institution on earth just admitted it by starting this small. I’d watch how many of those schools actually run the course next year. That number tells you far more than the 19 does.
We see people from universities all over the world, looking at our website and reading articles there every single day. If anyone at the university wants training we offer that and we have trained entire campuses of staff and professors and helped them develop AI training programs….JIC you don’t want to wait to get in at MIT.
Source: MIT Schwarzman College of Computing
California Now Requires Outside Audits of AI Systems
California signed two laws creating the first state system for independent AI audits. One sets up organizations that can verify AI systems, and the other builds a registry of approved AI auditors with rules on independence, due by January 1, 2029.
Both OpenAI and Anthropic backed the bills. If you use AI in hiring, lending, insurance, or any decision that affects a person, this is the framework that will eventually check your work.
My Take: OpenAI and Anthropic both supported these bills. When the biggest labs help write the rules they’ll be graded on, they aren’t being civic minded, they’re raising the cost of being their competitor. Audits are expensive. A giant lab absorbs that, and a six-person AI vendor doesn’t. Expect fewer AI suppliers in three years, not more, and expect the ones left standing to be whoever could afford the paperwork rather than whoever built the better thing. The registry isn’t due until January 1, 2029, which is the other tell. That’s a long runway to lobby.
As our business leader, you need to make sure that what you are building is easily auditable. Make sure your training information as well documented as well as your instructions. If you let it go two or three more years and you are subject to an audit that’s going to be really messy. You need to be able to pack up that digital suitcase and hand it over to them.
Source: Office of Governor Gavin Newsom
Anthropic Found AI Companies Secretly Sending Customer Questions to Claude
Anthropic published its latest report on people misusing Claude. It tracked about 40 groups and found attackers have moved on from single chats to full agent setups, with stolen API keys and sessions as the prize.
Three Chinese AI providers were passing their own customers’ questions through to Claude without telling them.
My Take: Everyone will file this under security. It’s a supply chain story. If a vendor can route your questions to a model you never chose and never mention it, the logo on your AI subscription tells you very little about what’s actually reading your work. The uncomfortable part is who caught it. Anthropic did, from its own side of the pipe. Not the customers, not a regulator, and not an audit, which means you have almost no way to find this yourself. The question I’d put to every AI vendor right now, in writing: which models touch our data? A slow answer is the answer.
Source: Anthropic
OpenAI Opened Its Agent Tools to Every Developer
OpenAI put its Agents API into public beta for all developers. It handles the messy parts of running agents: long tasks, agents that call other agents, and sandboxes to run them in. There’s no extra platform fee beyond the tokens and tools you already pay for.
Two early users reported real numbers. SafetyKit cut cost per case by 60%, and Hypha cut failed responses by 86%.
My Take: No platform fee isn’t generosity, it’s positioning. OpenAI is giving away the plumbing so your agents live on its models, and once your workflows are wired into its orchestration, leaving costs far more than the fee ever would have. Those 60% and 86% numbers are real, and they came from companies with engineers on staff, which is the part that won’t transfer to you. An agent that can run for a long time can spend for a long time. I wouldn’t point one at a paid API without a hard ceiling on it.
Source: OpenAI
Tool picks of the week
- Read AI: An AI notetaker that records your online meetings and writes the notes live.
- Snitcher: Identifies the companies visiting your website, reveals contact details, and triggers outreach automations.
- HeyGen: Creates studio-quality videos with lifelike avatars and voices for marketing and social media.
Got a burning question, a fresh take, or just want to share your latest AI wins? Hit us up at [email protected]. Your insights keep this community growing and thriving.
Have a great day!
-Nicole A. Donnelly
Founder, AI Smart Ventures
AI Strategy – AI Training – AI Consulting – AI Implementation
