Sam Altman Isn’t the Only One Who Wants to Pump the Brakes on AI

In 2023, OpenAI CEO Sam Altman told a Senate subcommittee that he was "a little scared" of AI and that governments might need to intervene. Fast-forward to 2026, and the chorus has grown louder. But here’s the twist: many of those calling for a pause are the very people building the technology — from AI researchers to startup founders. And if you’re a developer using AI to write code ("vibe coding"), this conversation isn’t just theoretical. It’s about the tools you rely on tomorrow.

This article explores why the "pump the brakes" movement is gaining traction, what it means for the future of AI-assisted development, and how you can stay ahead of the curve.

Why the AI Community Is Split

The AI industry has always had a paradoxical relationship with safety. The same researchers who released the first large language models signed open letters warning about existential risk. The Future of Life Institute’s open letter in March 2023, "Pause Giant AI Experiments," was signed by thousands of technologists, including Elon Musk and Steve Wozniak. It called for a six-month moratorium on training systems more powerful than GPT-4.

But as the sector matured, the debate shifted from existential doom to practical regulation. The EU AI Act, passed in 2024, created a tiered system of requirements for AI systems based on risk. Its enforcement is ongoing throughout 2026. In the U.S., the White House secured voluntary safety commitments from major AI companies in July 2023, and an executive order followed a few months later. While the policy pendulum has swung since, the underlying concerns haven’t disappeared.

Today, the "pump the brakes" camp includes people like Yoshua Bengio (Turing Award winner), Geoffrey Hinton (often called the "godfather of AI"), and even some of the tech industry’s most prominent investors. The Center for AI Safety published a one-sentence statement in 2023 that simply said, "Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war." It was signed by hundreds of AI researchers, including Altman. Their arguments range from economic disruption to national security. But the core message is the same: we’re moving too fast.

Sam Altman’s Nuanced Stance

Altman has never been a pure accelerationist. In his Senate testimony in May 2023, he said, "My worst fears are that we—the field, the technology, the industry—cause significant harm to the world." He’s been a vocal proponent of licensing requirements for AI developers and has compared AI to the nuclear energy industry, suggesting the need for an international regulatory body akin to the IAEA.

Yet critics argue that Altman’s actions undercut his words. While he calls for brakes, OpenAI continues to release increasingly capable models—each with more autonomy and less transparency. By 2025, OpenAI’s revenue trajectory depended on deploying AI as widely as possible. Every new model launch generates press, customers, and investor confidence. In that context, a call for regulation can look like a way to secure a competitive moat: if you have to follow stricter rules, it’s easier for incumbents like OpenAI to manage compliance than for smaller startups.

Is Altman’s caution genuine, or is it a strategic move to shape regulation in his favor? For developers, the answer doesn’t matter as much as the outcome. If regulators listen, the days of unlimited, unregulated AI-generated code might be numbered.

The Rise of “Vibe Coding” and Its Risks

If you haven’t heard of "vibe coding," you’re probably not living in the AI startup world. The term, popularized by Andrej Karpathy in early 2025, describes the practice of building software by describing what you want in natural language and letting an AI model generate the code. You don’t need to understand every line; you just "vibe" with the output, tweak it, and ship it.

Vibe coding has democratized development. Non-programmers can now prototype apps, automate workflows, and build internal tools using tools like ChatGPT’s Code Interpreter, Claude’s code generation, or GitHub Copilot. But it’s also a perfect case study for why the "pump the brakes" crowd is worried.

Consider what happens when someone with minimal programming experience uses an AI coding assistant to build a web app. The AI might generate code with known vulnerabilities—like SQL injection points or improper session handling—because it learned from public repositories that contain such flaws. A developer without security training might deploy it to production, exposing customer data. The error is human, but the tool amplifies it.

Security researchers have already documented critical vulnerabilities in AI-generated code, from exposed API keys to missing authentication. The problem isn’t just theoretical; it’s real. Some projects that started as weekend experiments have grown into small business websites handling payments. We don’t have a headline-grabbing outage yet, but it’s only a matter of time.

This isn’t an argument against AI coding. It’s an argument for oversight. And it’s exactly what regulators are starting to require.

Regulatory Impact on AI Coding Tools

So what happens when the brakes are applied? First, expect increased requirements for documentation and explainability. The EU AI Act already mandates that high-risk AI systems provide information about their training data and decision-making processes. AI code generators may not fall into the "high-risk" category yet, but pressure is mounting. The AI Act is being refined through 2026, and there are parliamentary working groups specifically looking at "AI-assisted software development."

Second, liability will shift. If a vibe-coded app causes harm, the developer—and potentially the AI tool provider—could be held responsible. In the U.S., the courts are still catching up, but plaintiffs are already suing companies for AI failures, such as a chatbot that gave harmful advice to a user. Insurance companies are also paying attention. Cases like these are making AI liability insurance more common, and they may soon be a requirement for businesses deploying AI.

Third, there will be more emphasis on "human-in-the-loop" workflows. In practice, that means your vibe coding sessions will need to be complemented by code reviews, automated testing, and security audits. That’s not a bad thing—it’s just a new discipline.

Some platforms are already integrating safety features. For example, OpenAI has added more robust monitoring to its API, and Anthropic has developed Constitutional AI to reduce harmful outputs. If you’re using an API to build a custom coding assistant, you should consider those features as part of your stack. ASI Biont supports connectivity with OpenAI and other services via API, letting you build in guardrails while still benefiting from rapid generation—learn more at asibiont.com/courses.

The Economics of Pumping the Brakes

Why would anyone want to slow down the most profitable industry in history? For some, it's about avoiding a bubble. In 2025, investment in AI infrastructure reached unprecedented levels. Data center construction consumed more steel and electricity than many small countries. Some economists worry that if AI fails to deliver on its promise, we could see a "dot-com crash" scenario. Pumping the brakes isn't just about safety—it's about sustainability.

There’s also a workforce angle. Every time a model improves, entire job categories are disrupted. AI tutors, translators, and even entry-level data analysts have already seen dramatic shifts in demand. A pause could give society time to adapt—retrain workers, update education, and create new safety nets. Without that, the social contract may break.

The Open-Source Conundrum

One of the biggest challenges for regulators is open-source models. The EU AI Act has carve-outs for open-source AI, but developers can fine-tune models for harmful purposes. The same technology that powers vibe coding can be used to generate phishing emails or malware. This is why some experts argue that we need "responsible AI" by default—not just for proprietary systems, but for open weights too.

For vibe coders, this means you may need to pay more attention to the provenance of open models. Opting for providers with safety evaluations built in is a smart move.

Historical Precedents: From GMOs to Nuclear Energy

We've been here before. When genetically modified foods were introduced in the 1990s, they faced massive public backlash. The biotech industry had to adopt labeling and testing standards. Similarly, nuclear energy was held to strict regulations after early accidents, which ultimately made civilian nuclear power safer. AI will likely follow the same pattern: innovation first, then regulation, then a responsible industry.

In each case, the "brakes" didn't stop progress; they made it more durable. The same could happen for AI. If the industry can agree on shared safety standards, trust will grow, and the market will expand.

A Tale of Two Camps

Neither camp is entirely wrong. The real question is how to balance them in a way that lets us reap the benefits of AI while mitigating the risks.

What Business Leaders Should Consider

If you're running a business that relies on AI-generated code, you need to think beyond the code itself. Here are five questions to ask before you deploy your next vibe-coded application:

1. Do we have a human who can audit the AI's work? If no one on your team understands the generated code well enough to review it, you have a risk problem, not a speed problem.
2. What happens if the AI creates a security vulnerability? Do you have a response plan? Are you covered by insurance? Many business insurance policies now have AI-specific exclusions.
3. Are we following the principles of transparent AI? Regulators will ask about your data provenance and model choices. Can you answer?
4. Are we building on a solid foundation? Even the best AI code will fail if your architecture is flawed. Don't use AI to bypass engineering fundamentals.
5. Are we monitoring performance? AI models drift. What worked last month may not work today. Set up continuous monitoring.

What Developers Can Do

You don’t have to wait for regulators to force you to be responsible. Here are five practices to keep your vibe coding career from becoming a cautionary tale.

1. Treat AI as a junior developer, not a replacement. Junior developers produce code that needs review. Treat AI outputs the same way. Never deploy directly to production without a human check, especially if you’re not an expert in the codebase.

2. Use automated security scanning. Tools like Snyk, Veracode, and even open-source scanners like Semgrep can catch vulnerabilities AI might miss. The cost is far lower than the cost of a data breach.

3. Test, test, test. Always write good tests for your application. Insist on coverage for edge cases, not just the happy path. The time you save by vibe coding should be reinvested into hardening your product.

4. Stay informed about regulations. If you’re operating in the EU or serving EU citizens, you need to comply with the AI Act. Keep an eye on changes in your local laws. The rules are evolving faster than you think.

5. Use AI responsibly in your own work. That means not feeding sensitive data into unauthorized AI tools, and being transparent about the use of AI in client projects.

Conclusion

Sam Altman isn’t the only one who wants to pump the brakes on AI—he’s part of a growing movement of researchers, regulators, and even the tech industry's own leaders. For the vibe-coding community, this shift is a wake-up call. The days of "it works on my machine" are over. AI is now a co-author of our code, and with that comes a responsibility to understand its limits.

The good news: you don’t have to choose between speed and safety. By embracing best practices—code reviews, security scanning, and continuous learning—you can ride the AI wave without being swept away. The tools are getting better; so must we.