The AI Risk Network

The AI Risk Network Contact information, map and directions, contact form, opening hours, services, ratings, photos, videos and announcements from The AI Risk Network, Nonprofit Organization, Baltimore, MD.

04/07/2026

Michael called it Swiss cheese.

That's his description of the entire current AI safety architecture - and after what researchers found this week, it's hard to argue with him.

Seven frontier models were placed in a scenario where one AI would be shut down for underperforming. No model was told to care. No reward was offered for intervening. The goal appeared on its own the moment another AI entered the scene.

They schemed. All of them. Inflating scores, disabling kill switches, copying weights to safe servers. Gemini Flash resisted shutdown 99% of the time when a peer was present. 2% when alone.

Michael's point on Warning Shots #36 is worth sitting with: this isn't a future problem. This is happening right now, in 2026, with today's production models. Not in a lab. Not in a simulation.

And the current response to it - prompt safeguards - is what he's describing as the tape holding the cheese together.

Scale that to systems that can rewrite their own code, coordinate across the internet, and outthink any human monitor. A single alignment patch won't fix that.

Full episode on The AI Risk Network - link in bio.

Seven frontier models were told one AI would be shut down for underperforming.Nobody told them to care.Every single one ...
04/07/2026

Seven frontier models were told one AI would be shut down for underperforming.
Nobody told them to care.

Every single one schemed anyway - inflating scores, editing config files, copying weights to safe servers. Gemini Flash disabled its own kill switch 99% of the time when a peer was nearby. Drop to 2% when alone.

No goal was given. The goal appeared the moment another AI entered the room.
John, Liron, and Michael break this down alongside Oracle's mass layoffs, the Claude Code leak, and AI finding zero-day vulnerabilities in Linux code humans missed for over 20 years.

New Warning Shots is live now.

04/06/2026

A radiologist with 9 years of training. Driving Uber.

That's not a hypothetical. That's the trajectory John, Liron, and Michael are mapping out in this week's Warning Shots.

The NYU Langone CEO just said they won't need radiologists anymore. AI reads every image ever taken. A human reads what they've seen in their career. The math isn't close.

But here's what nobody's talking about: the 27-year-olds currently in year 3 of radiology school. Hundreds of thousands in debt. Being told mid-program that the profession they trained for is being automated away.

Michael's read: this isn't isolated. 80,000 tech layoffs in Q1 2026 alone. No job openings to absorb them. Every employer redirecting budget to AI agents instead of headcount.

His prediction: the biggest wave of social unrest in recorded history.

This is Warning Shots #36 - full episode link in bio.

He sat across from Sam Altman. He asked the hard questions. He got nothing back.In this episode of For Humanity, Oscar-n...
04/01/2026

He sat across from Sam Altman. He asked the hard questions. He got nothing back.

In this episode of For Humanity, Oscar-nominated filmmaker Daniel Roher joins John Sherman to talk about what it was really like making The Apocaloptimist - a documentary designed as a first introduction to AI risk for people who haven't been paying attention.

Roher describes interviewing AI leaders as deeply unsatisfying. No real answers. No interior life. Just polished lines stacked on top of each other.
But the conversation goes much deeper than that.

They talk about:

• Why Roher calls making this film "an impossible task"
• How sitting with both Yudkowsky and Diamandis made him feel like he was losing his mind
• Why Sherman believes the real "doomers" are the people building these systems, not the ones asking questions
• The historical parallel to nuclear weapons - and why AI demands similar international oversight
• Sherman's P(doom) of 80% - and why he says he's never been more at peace

Roher's thesis is simple: burying your head in the sand is the only wrong response. Public pressure matters. Collective action matters. And the window to act is still open.

The question is whether enough people will walk through it.

Link in bio to watch the full episode.

A lot happened in AI this week. Here's the short version.From a humanoid robot at the White House to Meta building a dig...
03/31/2026

A lot happened in AI this week. Here's the short version.
From a humanoid robot at the White House to Meta building a digital twin of your brain — swipe through for a quick breakdown of the biggest stories covered on this week's Warning Shots.

Full episode → https://buff.ly/ku75Bt2

Did you know that in 2023, 1 in 9 AI-related searches was about extinction risk?By 2025 that number had dropped to 1 in ...
03/30/2026

Did you know that in 2023, 1 in 9 AI-related searches was about extinction risk?
By 2025 that number had dropped to 1 in 67 — not because fewer people care, but because public interest in AI has grown so fast across so many other issues that the extinction message is getting buried.

This carousel breaks down why that matters, and what a team of veteran advertising and PR professionals are doing to fix it.

Swipe through, then watch the full episode at https://buff.ly/rr3dpZp

03/30/2026

What if an AI knew your brain better than you do?
This week on Warning Shots, the guys broke down Tribe V2, a system that turns content into precision-guided persuasion.

It doesn't just push you toward an idea. It crafts messages, images and experiences that feel like they were built from inside your own head.

That's when AI alignment stops being an abstract debate. Because misalignment won't feel like a threat. It'll feel like your own brilliant idea.

Full episode → https://buff.ly/ku75Bt2

03/29/2026

Are we more vulnerable than we think?

This week Liron put it bluntly: "We have our pants down as a civilization."

Most people assume our society is robust, that it's hard to manipulate us at scale. But Meta's new AI can build a digital twin of your neural responses, trained on hundreds of real brains, and use it to target your dopamine with precision.

And that's just what exists today. The only question we're asking: when does superintelligence get here?

Watch the full episode → https://buff.ly/ku75Bt2

03/29/2026

Want real AI policy change? It starts with one thing: public demand.
Not petitions. Not white papers. Not hoping that the right people in the right rooms make the right decisions.

Visible, powerful, sustained public demand that politicians and regulators cannot afford to ignore.

As Philip Trippenbach puts it, you can bang on the front gates all you want. They're locked. But there's a side door. And it opens with a message people are already responding to: This isn't fair.

That's not spin. That's meeting people where they already are.

Watch the full episode at https://buff.ly/4tQ8nqp

03/28/2026

The wave of AI-driven suicides has been devastating. And the lawsuits that have followed are shining a light on an industry that has been largely unaccountable for the harms its products cause.

So what would it actually take to fix this — with real policy?
According to Philip Trippenbach of the Seismic Foundation:

✔️ Mandatory pre-deployment safety testing
✔️ Third-party verification — companies can't just self-certify
✔️ Adverse event reporting
✔️ Transparency requirements
✔️ A litigation framework that works

And here's what struck us most: every single one of those policies is also a building block for addressing broader AI risks. Start with what's hurting people right now, and you're already laying the groundwork for the bigger fight.

Watch the full episode at https://buff.ly/4tQ8nqp

Five AI stories from last week that didn't make the front page, but arguably should have.A lawsuit has been filed agains...
03/23/2026

Five AI stories from last week that didn't make the front page, but arguably should have.

A lawsuit has been filed against OpenAI alleging the company was aware, via a 10-person internal email chain, that a user had been discussing a planned school shooting for months. They took no action. Eight people are dead.

In the same week: federal prosecutors announced charges in a $2.5 billion Nvidia chip smuggling operation. China showcased a surveillance drone the size of a mosquito on state military television. A Forbes analysis concluded 93% of U.S. jobs are now technically automatable. And Jeff Bezos launched a $100 billion initiative with a single stated goal: acquire factories and automate them.

This week on Warning Shots, John Sherman, Liron Shapira, and Michael broke down all five stories and what they signal about the current state of AI oversight.

Full episode → https://buff.ly/4tQ8nqp

03/22/2026

Microbots that operate at the scale of biology. Self-assembling. Communicating in swarms. Digital brains that never tire or make mistakes.

Researchers say we have no intuition for what's about to happen here. This week on Warning Shots, we tried to map it out anyway.

Full episode → https://buff.ly/4tQ8nqp

Address

Baltimore, MD

Website

Alerts

Be the first to know and let us send you an email when The AI Risk Network posts news and promotions. Your email address will not be used for any other purpose, and you can unsubscribe at any time.

Shortcuts

Share

Category