OpenAI’s new model triggered California’s AI watchdog, the dispute reveals deeper tensions

Is OpenAI in trouble with the law?

Imagine building something amazing, only to be told you broke the rules while doing it. That’s the situation OpenAI faces right now. A watchdog group called The Midas Project claims the company violated a brand-new California safety law when it released its latest coding tool, GPT-5.3-Codex.

This isn’t just a small disagreement. If the group is right, OpenAI could be looking at serious fines. It’s a major test for California’s attempt to keep artificial intelligence in check, and everyone in the tech world is watching to see what happens next.

Meet the coding whiz causing all the fuss

GPT-5.3-Codex is OpenAI’s newest and most capable coding-focused model. It significantly outperforms previous OpenAI models on benchmarks like SWE-Bench Pro and Terminal-Bench 2.0 and achieves state-of-the-art results on several coding and “agentic” tests.

The company says it beats all its older models and even beats competitors like Anthropic at coding challenges. It’s a big deal for OpenAI because they want to stay the king of the hill in AI-powered coding.

What is California’s new AI safety law?

California recently passed a law called SB 53, and it’s a game-changer for artificial intelligence. The law basically says that big AI companies have to pick a safety plan and actually stick to it. They can’t just make promises; they have to follow through with real actions.

The goal is to prevent catastrophic risks, things that could cause a lot of harm, like major cyberattacks. It also makes it illegal for companies to lie about following their own rules. It’s one of the first laws of its kind in the U.S., and it sets a very high bar for how AI should be handled responsibly.

Why this model is called high risk

OpenAI uses its own internal system to grade the danger level of its creations. It’s called the Preparedness Framework. For the first time, a model, GPT-5.3-Codex, hit the high-risk category for cybersecurity.

That might sound scary, but what does it mean? Basically, the model is so good at coding that it could potentially be used to cause serious digital harm, especially if someone used it to automate cyberattacks.

The watchdog’s shocking accusation

Here’s where the problem starts. The Midas Project, the watchdog group, says OpenAI did not put those extra safety measures in place before releasing the model. They claim the company dropped the ball on safety.

According to the group, OpenAI’s safety plan requires special safeguards when a model is this powerful. These safeguards are designed to stop the AI from doing things like acting sneaky or hiding its true abilities from researchers.

OpenAI fires back, we followed the rules

Of course, OpenAI isn’t just taking this accusation quietly. They strongly disagree with the watchdog’s claims. A company spokesperson told Fortune magazine that they are confident they followed the law properly.

Their defense is based on a technicality. They argue that the extra safety rules only kick in when a model has both high cyber risk and something called long-range autonomy, the ability to work on its own for a long time.

The debate over long range autonomy

OpenAI acknowledges that it uses proxy evaluations rather than a perfect metric to gauge long-range autonomy.

Critics note that earlier Codex models already set state-of-the-art results on long-horizon benchmarks and could autonomously complete complex tasks, making it difficult to accept that the newer GPT-5.3-Codex falls below any meaningful autonomy threshold.

They argue that if OpenAI found the criteria unclear, it should have clarified them with regulators before launching such a powerful model.

A founder’s frustration, this is embarrassing

Tyler Johnston, the founder of The Midas Project, didn’t hold back his feelings about the situation. He said the potential violation is especially embarrassing because the law doesn’t ask for much from companies.

In his view, the law just requires companies to pick a safety plan they like and then be honest about following it. You can even change the plan if you need to along the way. His point is that the bar is set pretty low, so failing to meet it is a big deal.

Other experts weigh in on the dispute

It’s not just the Midas Project speaking up. Other safety researchers are also questioning OpenAI’s defense. Nathan Calvin, who works for a group called Encode, said he read all the relevant documents, and the rules didn’t seem confusing to him at all.

When independent experts start agreeing that the rules look clear, it puts OpenAI in a tougher spot. It makes their ambiguous language defense seem a little weaker. The tech community is now split, with some taking the company’s side and others wondering if they really did try to bend the rules to get ahead.

What happens if they’re found guilty?

So, what’s the worst that could happen to OpenAI? If an investigation is opened and the accusations are proven true, the California Attorney General’s office can step in to enforce the law. Under the new law, the penalties can be steep.

We’re talking about fines that could reach millions of dollars, depending on how serious the violation was and how long it went on. A spokesperson for the Attorney General said they are committed to enforcing AI laws, though they wouldn’t say if an investigation into OpenAI is happening.

Business is booming despite the drama

Even with all this legal drama, OpenAI is doing incredibly well as a business. CEO Sam Altman recently told his employees that ChatGPT is growing fast again, with monthly users jumping more than 10 percent recently.

He also had some good news about the very model at the center of the controversy. Usage of Codex shot up by about 50 percent right after GPT-5.3-Codex was released, especially with a new Mac app. So, while the lawyers might be worried about fines, the customers seem to love the product and its capabilities.

A side battle with rival Anthropic

The tension with regulators isn’t the only thing on OpenAI’s plate. They are also in a heated rivalry with another AI company called Anthropic. Things got personal recently over Super Bowl ads that ran during the big game.

OpenAI’s rivalry with Anthropic has spilled over into marketing. Anthropic has promoted its models with high-profile campaigns, including Super Bowl-bound ads that take aim at OpenAI’s plans to introduce ads into the free version of ChatGPT.

Sam Altman publicly criticized these ads and defended OpenAI’s commitment to keeping any advertising clearly separated and transparent, while the company experiments with small, clearly labeled ad units near the bottom of some responses.

And if you’re wondering what else OpenAI is building behind the scenes, take a look at how it’s introducing age prediction tech to ChatGPT.

A major test for AI rules everywhere

This whole situation is bigger than just one company or one model. It’s the first major test of whether new AI laws can actually work in the real world. Can regulators keep up with technology that changes every single month?

The outcome will set a precedent for the entire industry. If OpenAI, a giant in the field, can be found in violation, it sends a clear message: no one is above the law. It proves that states are serious about enforcing safety rules, and it might even push the U.S. toward adopting a single, clear set of national standards for artificial intelligence.

And if you want to see how the pressure around OpenAI is escalating on other fronts too, take a look at how its clash with Elon Musk just got bigger.

If you found this story about AI and the law interesting, give it a thumbs up and drop a comment below. We’d love to hear your take on whether OpenAI should face those big fines.

This slideshow was made with AI assistance and human editing.

Don’t forget to follow us for more exclusive content on MSN.

Read More From This Brand:

• Why OpenAI’s AI music generator could be a step too far?
• OpenAI identifies mental health concerns among ChatGPT users
• OpenAI’s ChatGPT might lose safety guardrails after long-term use