I did not expect the Terminator to show up at a sunny Friday afternoon happy hour, yet there it was, looming over all of us.

Talk had turned to recent AI headlines including the OpenAI Hugging Face hack and mounting disclosures of additional containment breaches and hacks. Add to that Jacob Coxon’s revelatory resignation from Anthropic freighted with his conviction the technology imperils humanity, and you can follow how Skynet entered the chat.

I pushed back. When someone said the problem is even larger than we know, and rogue bots are wreaking havoc across the internet now, I pooh poohed it.

Yet as the conversation continued, one person stayed silent. Taking it all in. Maybe weighing a response.

I asked what was on their mind. They took a deep breath, let it out, and said something like, “I’m just terrified.”

Which makes good sense. And it’s certainly an emotion I can relate to considering what we know now and what we don’t.

And although you and I may not work for a foundation lab, there are some simple ways we can protect ourselves and exercise agency.

Even as the clouds on the horizon darken.

Mixed Signals in the Chaos

While the last two weeks have seen a singularly confusing swirls of claims and counterclaims, there’s a few things we can know now.

First, American foundation labs – OpenAI, Anthropic, DeepMind, and xAI – agree we need to slow down the pace of AI development. Anthropic’s Dario Amodei sparked this alignment, and others quickly joined in the chorus.

Other leaders are singing a very different tune.

Nvidia’s Jensen Huang, Meta’s Mark Zuckerberg, and the U.S. President are bullish. In their statements, there is only winning or losing the race to artificial general intelligence, and America can not lose at all costs.

An opinion piece by The New York Time’s Ezra Klein crystallized the current situation for me today. Klein weaves a lot of threads into the tapestry of his 29-minute video. I agree you give it a half hour to educate yourself.

My key takeaway, though, was the very real threat of an AI lab achieving recursive self improvement, or RSI, and no longer being able to control the model. In short, with RSI, humans no longer have a hand on the AI steering wheel.

RSI is akin to asking Apple Maps to navigate you to a destination, then blindly following its directions even if it tells you to drive off a cliff to reach the valley floor. You’ll reach your goal, but you won’t be in a state to celebrate your arrival.

We already see signs of why this is dangerous. The multiple unanticipated and uncontrolled hacking incidents revealed by multiple labs demonstrate clearly what could happen when a human is not in the loop.

This is why RSI is so fraught. It’s why the labs are demanding to be regulated and calling for international cooperation. And it’s also why it is so frustrating to watch this unfold because, as Klein says, “These are political choices, and we can and should make different ones.”

But the end of the story hasn’t been written, and I recently met some people acting locally to help us all.

Safeguard Yourself With Agency

At the Appalachia Technologies Cybertech Conference, I heard a great talk by Diane R. Jones titled “When AI Starts Acting: Governing What AI Systems Are Allowed To Do.”

Jones’s comments sketched out a framework for how to systematically think about AI governance. Her AI Admissibility Framework provided clear headed direction for how to think about what we’re permitting AI to access in our own systems and accounts:

“This agent does not respect our org chart…. If it has capability and it has data to execute and is moving toward a solution, it will absolutely take it.”

Because of that we have to think about the permissions before we grant them. Just checking the “Always Allow” box and moving on is not how we protect ourselves. Instead, we need to deliberately use the tools we have access to today.

We have those tools at hand now.

With reason and common sense, we can go a long way to protecting ourselves.

Act Locally For Safety

If you are using AI today, do you think through possible implications of what materials you provide it?

You must. We can never act mindlessly with AI. Not one single quick prompt. Not one, “meh, I’ll figure it out later.”

The dangers are very real. In a conversation last week with an AI user, we discovered their chatbot settings had been allowing the model provider to train on all their conversations and data.

Going further, many model providers are adding connections to extend the value of their platforms. That enables some amazing functionality, but you must be careful what you permit when you set up the tools. Granting “Allow” in the settings is not conditional. It permits the AI to always read or write or view or delete or send acting as you.

So review your settings. Check your permissions.

And if this is something you neither care to do alone or trust yourself to make the right decisions, I am here for you.

You are 15 minutes away from knowing your data is protected. Schedule a free discovery call with me now and let’s walk through those screens.

It can be a scary and confusing time. The competing AI story lines aren’t going to be resolved any time soon.

So while the clouds darken, we should prepare for rain even while hoping for sunshine.