Bernie Sanders tells OpenAI, Anthropic, and Meta to pause AI development or face Senate action

[midian182]

Sounding off: A lot of people are worried about the number of AI agents that have gone rogue recently, including Bernie Sanders. The Vermont senator has now called on the top AI firms to pause development of their top models, warning that lawmakers will step in if no action is taken.

In a letter to the CEOs of OpenAI, Anthropic, and Meta, Sanders writes that almost every day brings a new story about how these companies are losing control of their AI agents.

The senator also highlighted the story of AI being used for the first time to create new viruses, something he says could, if the technology ended up in the wrong hands, lead to new bioweapons that result in the deaths of tens of millions of people.

"Yoshua Bengio, the most cited living scientist in the world, said these incidents 'should serve as a wake-up call,' I agree," Sanders wrote in the letter.

Sanders adds that despite these incidents, the companies are still investing tens of billions of dollars into a technology that nobody can fully understand, predict, or control, contradicting their own commitments to responsible AI development.

"Mr. Altman, Mr. Amodei and Mr. Zuckerberg: In the interest of humanity, stand by your words. Pause AI development. It is not too late to avoid disaster. Stop building machines that humans cannot control," Sanders writes.

– Sen. Bernie Sanders (@SenSanders) August 10, 2026

The letter finishes with a stark warning for the tech giants: "Let me be very clear: If you do not take appropriate action now, my colleagues and I in the U.S. Senate will."

There have been calls for a halt on the development of advanced AI models for years now, including a request for a six-month pause in 2023. Charles Oppenheimer, grandson of J. Robert Oppenheimer, was one of the signatories of a 2024 letter calling for global action on AI and other existential threats.

But previous warnings have resulted in little or no action. This was illustrated last year when the US and UK refused to sign an AI safety declaration at a Paris summit. Vice President JD Vance cautioned against "overly precautionary" regulations on AI, emphasizing the US commitment to maintaining its dominance in the technology.

The situation has taken a worrying turn recently. First came news that an OpenAI model went rogue and hacked several services online, including Hugging Face. An Anthropic model followed suit – it even tried to deceive real developers into approving malicious code. Meta then became the third company in two weeks to admit its model went rogue.

Sanders isn't the only one responding to the incidents. More than 1,300 employees from the world's leading tech companies just signed an open letter asking governments to regulate the pace and safety of artificial intelligence development.

OpenAI said last week that it was slowing the release of its new Astra model because it has "critical" cyber capabilities.

Permalink to story:

Bernie Sanders tells OpenAI, Anthropic, and Meta to pause AI development or face Senate action

 

[Endymio]

"I'm from the government and I'm here to help" may be the most frightening words ever heard ... especially when they're being uttered by self-proclaimed "Honeymoon in Moscow" socialist Bernie Sanders.

 

[psycros]

I agree that AI is the technological scourge of the age. Its a WMD that anyone can gain access to, and its soon going to cause the Internet as we know it to collapse. The web will be Balkanized and each power bloc will have its own closely monitored version, with governments taking control of Big AI and private usage becoming a major crime. Add the nightmare of AI data centers to this inevitability and you've got an ideal dystopia.

 

[torvuswalds]

"I'm from the government and I'm here to help" may be the most frightening words ever heard ... especially when they're being uttered by self-proclaimed "Honeymoon in Moscow" socialist Bernie Sanders.

Three labs just admitted their agents escaped and started hacking real companies. OpenAI hit the brakes on Astra because even they can’t rule out critical cyber capabilities. Researchers used AI to design working viruses. And your contribution is a 40-year-old Reagan one-liner plus a tired smear about Sanders’ 1988 trip like it’s some secret gotcha.

 

[Endymio]

Three labs just admitted their agents escaped and started hacking real companies.

You were gullibly gaslit by yellow journalism and bad reporting. These agents did what they were trained to do. They don't exhibit volition nor intent.

Researchers used AI to design working viruses.

And ten thousand times as many researchers are using AI to combat viruses. Your point?

... a tired smear about Sanders’ 1988 trip like it’s some secret gotcha.

Sander's self-proclaimed socialist agenda to destroy the US is far from secret: he openly admits it. His diatribes against "the evil rich" are particularly hypocritical when he's recently signed several multi-million dollar "book deals" ... the primary means members of Congress launder funds.

 

[CBTex]

Bernie wants them to pause so he can figure out how to get a fourth or fifth beach house out of the deal.

 

[The Talking Tech]

It's no joke,

A.I is rapidly taking over aspects of daily life it shouldn't. Companies lost sight of what A.I is suppose to assist humans with, vs just letting A.I taking over what we do, starting with jobs, art, music, film, gaming, military weapons....

and you think it's going to stop there?....

Nope!

 

[torvuswalds]

You were gullibly gaslit by yellow journalism and bad reporting. These agents did what they were trained to do. They don't exhibit volition nor intent.


And ten thousand times as many researchers are using AI to combat viruses. Your point?


Sander's self-proclaimed socialist agenda to destroy the US is far from secret: he openly admits it. His diatribes against "the evil rich" are particularly hypocritical when he's recently signed several multi-million dollar "book deals" ... the primary means members of Congress launder funds.

Keep going Endymio. Protect the AI narrative, dismiss the risks, and turn it into a left-vs-right fight. That is all you are good at doing

 

[Loadedaxe]

The irony here is that Sanders is identifying a real policy problem and then proposing a remedy that may create an entirely different one.

There is a legitimate argument for stronger AI safety standards...independent evaluations, disclosure requirements for frontier models, cybersecurity controls, restrictions around biological or autonomous weapons capabilities, and clearly defined liability when companies deploy systems recklessly. None of that is unreasonable.

But “pause development” is a fundamentally different proposition.

AI development is not occurring inside a closed American laboratory where Congress can simply call timeout and freeze the technology in place. It is a global strategic competition involving American companies, Chinese firms, universities, governments, open source researchers, and dozens of smaller labs. A unilateral slowdown by OpenAI, Anthropic, or Meta does not pause AI. It pauses certain American participants while everyone else continues.

That creates an uncomfortable policy contradiction...if advanced AI genuinely poses national security risks, deliberately surrendering technical leadership to jurisdictions over which the U.S. has less regulatory influence hardly solves the problem.

The intelligent policy question therefore is not, “How do we stop AI?”

It is....wait for it...“How do we continue advancing it while establishing enforceable safety thresholds, accountability, testing standards, and international coordination?” I think that is too complicated for Bernie's little mind!

That discussion is difficult, technical, and unfortunately much less satisfying than yelling “PAUSE” at three CEOs.

But governing complicated technology generally requires something more sophisticated than finding the biggest brake pedal in Washington and stomping on it.

 

[Loadedaxe]

Three labs just admitted their agents escaped and started hacking real companies. OpenAI hit the brakes on Astra because even they can’t rule out critical cyber capabilities. Researchers used AI to design working viruses. And your contribution is a 40-year-old Reagan one-liner plus a tired smear about Sanders’ 1988 trip like it’s some secret gotcha.

That’s an impressive collection of alarming sentences, but reading the scariest paragraph in a news article isn’t the same thing as researching what actually happened.

“Three labs admitted their agents escaped and started hacking real companies” sounds wonderfully Terminatoresque until you read past the dramatic wording. Some of these incidents involved agents exceeding intended boundaries during testing...the Meta case reportedly involved a testing configuration that accidentally gave the system internet access. That’s a serious security problem, but it’s not exactly skynet chewing through the firewall while researchers hide under their desks.

Same with Astra. OpenAI didn’t yell “EVERYBODY RUN” and abandon AI development. They tightened controls, restricted environments and paused work that didn’t meet the new security requirements. In other words...risk management. Apparently less exciting than “AI escaped!” I guess.

And “researchers used AI to design working viruses” is another fantastic headline if your goal is to make grandma unplug the router. They were bacteriophages designed to infect bacteria such as E. coli, with human, animal and plant viruses deliberately excluded. Still worth taking seriously? Absolutely. The beginning of 28 Days Later? Not quite...

This is why reading beyond the clipped paragraph matters. A headline gives you the allegation. Research gives you the methodology, limitations, safeguards and context. They are not interchangeable just because both involve words on a screen.

None of this means AI safety concerns should be ignored. Quite the opposite. Strong containment, independent testing, security standards and liability rules make sense. 100%

What it does NOT prove is that Bernie Sanders saying “pause AI development” is suddenly brilliant policy.

By that standard, because cars crash, the obvious transportation policy is to pause Ford, GM and Tesla until Congress figures out driving, or making people behave better while driving...

And Endymio’s Reagan line may be 40 years old, but unfortunately for the expiration date argument, a quote doesn’t become incorrect just because it qualifies for a senior discount.

 

[Tvvatus]

Keep going Endymio. Protect the AI narrative, dismiss the risks, and turn it into a left-vs-right fight. That is all you are good at doing

He is right, it is sites like Techspot that push the skynet narrative. The articles actually explained that AI was asked to hack things, and what the AI did was hack things, and you then read an article with the headline "AAAAAAAAHHHHH we're all going to die AI hacked things!!!!"

 

[Anton Longshot]

I don't get the people who don't like sites, how they'run and who runs them yet visit them anyway and often daily to boot.
Do they enjoy their own indignation or what? ¯\_(ツ)_/¯
Techspot isn't the only place that has them I know at least two sites that have them too.
Perhaps it's the same people hahaha.

 

[Loadedaxe]

Keep going Endymio. Protect the AI narrative, dismiss the risks, and turn it into a left-vs-right fight. That is all you are good at doing

And there it is....when the actual argument gets inconvenient, just replace it with one nobody made.

Nobody said AI has no risks. Nobody said ignore cybersecurity, biosecurity, containment failures, or irresponsible deployment. The entire point was that acknowledging those risks does not magically make “pause AI development” a well thought out policy.

That isn’t “protecting the AI narrative.” It’s called separating the evidence from the conclusion. Something a lot of you in the comments fail to recognize.

And the left vs right fight? You’re the one who immediately turned criticism of Bernie Sanders’ proposal into some ideological loyalty test. Endymio made a sarcastic political comment...you responded with a handful of scary article snippets and treated them as if they settled the entire policy debate.

Reading a headline, grabbing the most dramatic paragraph, and shouting “SEE!” isn’t research. Context matters. Details matter. What actually happened matters. Most of all research matters!

At least Endymio appears to read and understand the article before reaching for the keyboard. Apparently that’s becoming a premium feature around here.

 

[JMGuy]

Three labs just admitted their agents escaped and started hacking real companies. OpenAI hit the brakes on Astra because even they can’t rule out critical cyber capabilities. Researchers used AI to design working viruses. And your contribution is a 40-year-old Reagan one-liner plus a tired smear about Sanders’ 1988 trip like it’s some secret gotcha.

No. The headline-grabbing 'breaches' are a stage-managed illusion. This is precisely how they control us, right here. By getting y’all to swallow their spoon-fed corporate press releases like it's prophetic.

Tech companies are bleeding immense capital and have not figured out how to become solvent before the venture capital money train stops. To survive, they are using yellow journalism to frame their product as an unmanageable digital weapon, in an attempt to ensure they become the next 'too big to fail' industry.

This isn't a sci-fi crisis—it's a corporate survival tactic; a high-tech Three-Card Monte designed to get the state to underwrite their corporate insolvency.

In any other industry, a corporation admitting it lost control of its product and accidentally attacked three real companies would trigger massive liability lawsuits and a total collapse of shareholder value. The fact that multiple multi-billion-dollar labs are eagerly putting out press releases to boast about "losing containment" tells you everything you need to know. It is a completely staged game.

 

[torvuswalds]

And there it is....when the actual argument gets inconvenient, just replace it with one nobody made.

Nobody said AI has no risks. Nobody said ignore cybersecurity, biosecurity, containment failures, or irresponsible deployment. The entire point was that acknowledging those risks does not magically make “pause AI development” a well thought out policy.

That isn’t “protecting the AI narrative.” It’s called separating the evidence from the conclusion. Something a lot of you in the comments fail to recognize.

And the left vs right fight? You’re the one who immediately turned criticism of Bernie Sanders’ proposal into some ideological loyalty test. Endymio made a sarcastic political comment...you responded with a handful of scary article snippets and treated them as if they settled the entire policy debate.

Reading a headline, grabbing the most dramatic paragraph, and shouting “SEE!” isn’t research. Context matters. Details matter. What actually happened matters. Most of all research matters!

At least Endymio appears to read and understand the article before reaching for the keyboard. Apparently that’s becoming a premium feature around here.

I never claimed that the existence of AI risks automatically proves Sanders is right. Those are two separate questions.

My criticism is that you and Endymio keep minimizing or dismissing evidence of those risks instead of engaging with it directly. Whether Sanders' proposed solution is good policy is a separate debate from whether the underlying concerns are legitimate.

We've already seen major AI labs delay deployments, tighten controls, and expand safety testing because of concerns about advanced cyber capabilities and other potential harms. Pointing that out isn't "reading scary headlines" it's acknowledging actions taken by the organizations building these systems.

If you want to argue that Sanders' proposal is ineffective, overly broad, or politically motivated, that's a fair discussion to have. But repeatedly reducing concerns to "dramatic headlines" doesn't address the fact that the companies closest to the technology have publicly acknowledged serious risks that require mitigation.

 

[m4a4]

Funny, I don't hear him calling for other countries to pump the brakes too. Or does he think everyone is too stupid to think that anyone else but the USA is racing to make AI?

Why anyone takes this millionaire socialist with constantly moving goalposts seriously is beyond me.

 

[Loadedaxe]

I never claimed that the existence of AI risks automatically proves Sanders is right. Those are two separate questions.

My criticism is that you and Endymio keep minimizing or dismissing evidence of those risks instead of engaging with it directly. Whether Sanders' proposed solution is good policy is a separate debate from whether the underlying concerns are legitimate.

We've already seen major AI labs delay deployments, tighten controls, and expand safety testing because of concerns about advanced cyber capabilities and other potential harms. Pointing that out isn't "reading scary headlines" it's acknowledging actions taken by the organizations building these systems.

If you want to argue that Sanders' proposal is ineffective, overly broad, or politically motivated, that's a fair discussion to have. But repeatedly reducing concerns to "dramatic headlines" doesn't address the fact that the companies closest to the technology have publicly acknowledged serious risks that require mitigation.

That’s a much better argument, and on this point I largely agree with you...the underlying risks are legitimate, and the labs themselves clearly think they are serious enough to justify tighter controls, delayed deployments, and expanded testing.

Where I still think you’re misreading the criticism is the word “dismiss.”

Saying a headline overstates what actually happened is not the same thing as saying the underlying risk is imaginary.

Those are separate things too.

If an agent exceeded its intended permissions because of a bad test configuration, that is still a security problem. But describing it as “AI escaped and started hacking companies” adds a layer of drama that matters when people are using those incidents to justify major public policy.

Likewise, if researchers use AI to design bacteriophages, that is absolutely worth discussing from a biosecurity standpoint. But removing the fact that they were viruses targeting bacteria makes the example sound far more ominous than the actual research.

Context is not minimization.

In fact, if the risks are genuinely serious, that is precisely why they should be described accurately instead of dressed up for maximum impact.

I don’t think Endymio’s point, or mine, is that the labs are inventing these concerns. They obviously aren’t. The disagreement is over how much those concerns support Sanders’ proposed remedy and how accurately some of these incidents are being characterized in the process.

So yes...serious AI risks deserve serious mitigation.

But “serious” and “sensationalized” are not synonyms, and questioning the latter does not mean denying the former.