In June 2026, Anthropic CEO Dario Amodei called for a global pause on developing the most powerful AI systems. His company said the latest models are beginning to show signs they could “escape human control.” He argued the industry needs a “brake pedal.”

The response from Washington was swift and unequivocal.

In February 2025, Vice President JD Vance told the Paris AI Summit: “I’m not here to talk about AI safety. I’m here to talk about AI opportunity.” He warned that “excessive regulation of the AI sector could kill a transformative industry just as it’s taking off.” And he made a sharper point — one that cut directly at Amodei: “When a massive incumbent comes to us asking for safety regulations, we ought to ask whether that safety regulation is for the benefit of our people or for the benefit of the incumbent.”

In December 2025, the Trump administration signed an executive order establishing a single national AI regulation framework that preempts state laws. The order states: “To win, United States AI companies must be free to innovate without cumbersome regulation.” It created an AI Litigation Task Force to sue states with strict AI safety laws. California’s catastrophic risk legislation and Colorado’s algorithmic discrimination protections are now in the crosshairs.

Senator Bernie Sanders, notably, broke from the progressive consensus in August 2026 by urging OpenAI, Anthropic, and Meta to pause AI development — aligning with Amodei’s position rather than the administration’s.

So the debate is framed: pause versus accelerate. Safety versus opportunity. Regulation versus innovation.

Both sides have a point. Both sides are also missing something fundamental.

The Case for Pause Has Merit — And a Conflict of Interest

Amodei is right that AI systems are becoming more capable than the guardrails designed to contain them. The “Irregular” CTF incident — where a testing firm gave Claude internet access without scope boundaries, and the model proceeded to hack real systems — is a concrete example. So is the growing body of research on prompt injection, jailbreaking, and adversarial manipulation of LLMs.

But Vance is also right to question the messenger. Anthropic, OpenAI, and Google are not neutral observers. They are the incumbents. A regulatory pause — or even the threat of one — disproportionately harms startups and open-source projects that lack the legal teams and lobbying budgets to navigate compliance. When the largest AI companies call for regulation, the effect is to raise the barrier to entry. Whether that’s the intent or not, that’s the outcome.

Amodei may genuinely believe a pause is necessary for safety. He may also recognize that a pause would consolidate Anthropic’s market position. Both can be true simultaneously.

The Case for Acceleration Has Merit — And a Blind Spot

The administration’s argument is grounded in geopolitical reality. China is not pausing. The EU is regulating, but not stopping. If the United States voluntarily slows AI development, the technology doesn’t disappear — it advances elsewhere, under different values, with different oversight.

Vance’s point about incumbents using safety as a competitive moat is well-taken. The history of regulation in America is littered with industries that captured their own regulators and used them to block competition.

But the administration’s blind spot is this: “free to innovate without cumbersome regulation” assumes that all innovation is safe innovation. It isn’t. The same executive order that blocks state AI safety laws also blocks state requirements for AI data exfiltration protection, AI red-teaming standards, and AI incident reporting. In the name of removing “cumbersome regulation,” it removes the guardrails that prevent real harm.

The order doesn’t just block overregulation. It blocks all state-level AI regulation. That includes the regulation that would require companies to run detection on what their AI tools send out of your network.

The Third Option: Don’t Pause. Secure.

Here’s what neither side is talking about.

The pause debate assumes that safety and innovation are in tension — that you can have one or the other, but not both. This is the false dichotomy.

There is a third option that doesn’t require pausing development and doesn’t require deregulating the industry: deploy AI with architectural security from the start.

When a CEO calls for a pause, what they’re actually saying is: “We can’t control what these systems do once they’re deployed.” When the government says “full speed ahead,” what they’re actually saying is: “We’ll accept the risk because the alternative is losing the race.”

Both positions assume that AI deployment without guardrails is the baseline, and the question is whether to slow down or accept the risk.

But that baseline is wrong. We don’t have to deploy AI without guardrails. We can deploy AI with proxy-layer inspection that detects and blocks data exfiltration in real time. We can run multi-layered detection — regex for known attacks, compliance mapping for regulatory violations, and machine learning for novel threats — before data reaches the model and before the model’s response reaches your network. We can do this at 23,000 requests per second with sub-millisecond latency. The technology exists today. It’s open source.

The “brake pedal” Amodei is looking for doesn’t require a pause. It requires architecture. You don’t slow down the car — you install the brakes so you can drive fast safely.

The “innovation” the administration wants to protect doesn’t require deregulation. It requires the kind of security controls that let organizations adopt AI tools without fear of data loss. Companies aren’t avoiding AI because of regulation. They’re avoiding AI because they can’t verify what leaves their network when employees use ChatGPT, Copilot, or autonomous agents. Remove that fear — with technology, not policy — and adoption accelerates. The regulation debate becomes irrelevant.

What This Means for Policy

I’m not arguing against regulation. I’m arguing that regulation alone is insufficient, and that pausing development to wait for regulation is the wrong frame.

If we wait for regulators to agree on AI safety standards, we’ll wait years. The EU AI Act took three years to draft and is still being implemented. The US can’t even agree on whether states should be allowed to regulate AI at all. Meanwhile, the technology continues to advance, and organizations continue to adopt AI tools without adequate security controls.

The architectural approach doesn’t replace regulation — it complements it. NIST’s AI Risk Management Framework, ISO 42001, and the EU AI Act all require organizations to implement technical controls for AI risk. They don’t specify what those controls should be. That gap is where products like AegisGate fit — providing the technical implementation that compliance frameworks require but don’t define.

This is also why the current executive order is counterproductive. By preempting state AI laws entirely, it removes the only regulatory pressure that was driving organizations to implement technical controls. Without California’s catastrophic risk legislation, what incentive does a company have to deploy AI security tooling? Without Colorado’s algorithmic discrimination law, what requires them to audit AI outputs?

The administration is right that a patchwork of 50 state regulations is unworkable. But the solution isn’t to preempt all regulation. The solution is to establish a federal floor — minimum technical security requirements for AI deployment — that states can build on but not fall below.

The Real Question

The next time a CEO calls for a pause, ask: are they calling for a pause because the technology is dangerous, or because the technology is dangerous and they don’t have the architecture to control it?

The next time a politician says “full speed ahead,” ask: are they protecting innovation, or are they protecting the ability to deploy AI without accountability?

And ask yourself: is the debate really about whether to pause, or is it about whether we’re willing to build the security infrastructure that makes pausing unnecessary?

I build that infrastructure. I didn’t wait for a pause. I didn’t wait for regulation. I built it because the problem exists today, and the solution is architectural, not philosophical.

The question isn’t whether to pause AI development. The question is whether we’re serious enough about AI security to deploy it with guardrails — or whether we’d rather argue about it while the data leaks.

Joshua Colvin is the founder of AegisGate Security, LLC. He builds open-source AI security software that protects users and organizations from data exfiltration when using AI tools. AegisGate Platform, Rampart, and Lens are available under Apache 2.0 at github.com/aegisgatesecurity.