Key Takeaways

• A proposed securities class action filed Friday in federal court in California alleges Alphabet misled investors about the expected launch of Google's Gemini 3.5 Pro AI model.
• The lawsuit alleges CEO Sundar Pichai publicly indicated the model was coming in June while Alphabet allegedly failed to disclose disappointing training results and a significant delay.
• Alphabet shares fell roughly 4.4% on July 16 after a report said Google was months behind schedule. The allegations are new, Alphabet has not yet responded in court, and no class has been certified.

CALIFORNIA - A newly filed federal securities lawsuit is raising a question that could become increasingly important as America's largest technology companies race to build more powerful artificial intelligence: How much must a company tell investors when a highly anticipated AI model isn't performing as expected behind closed doors?

On Friday, Oct. 2, investor Chris Stewart filed a proposed class action against Alphabet Inc., CEO Sundar Pichai and CFO Anat Ashkenazi in the U.S. District Court for the Northern District of California.

The case is Stewart v. Alphabet Inc., et al., No. 3:26-cv-11290.

Federal docket records show the complaint, proposed summons and corporate disclosure statement were all filed Friday. Docket Nexus

The lawsuit proposes representing investors who acquired Alphabet securities between May 19 and July 16, 2026. It alleges violations of Section 10(b) of the Securities Exchange Act and SEC Rule 10b-5, along with control-person claims under Section 20(a).

The allegations have not been proven, and the lawsuit is at its earliest stage.

But what makes this case different from an ordinary shareholder dispute is what allegedly happened inside Google's AI development process while investors were hearing optimistic statements about its next major model.

According to the complaint, Pichai discussed Gemini 3.5 Pro during Google's May 19 I/O developer conference and indicated that the model would arrive the following month. During a June 3 investor call, he again allegedly said Alphabet expected Gemini 3.5 Pro to arrive in June. Open Class Actions

The lawsuit alleges investors weren't getting the whole picture.

Specifically, Stewart claims Alphabet failed to disclose that Gemini 3.5 Pro was producing disappointing training results, that those problems would significantly delay its launch and that the company's positive statements consequently were misleading or lacked a reasonable basis.

That allegation goes directly to the heart of federal securities law.

Companies don't necessarily have to disclose every internal problem, engineering setback or unsuccessful experiment. AI development in particular can involve models that improve, regress or fail during training.

But once a public company chooses to speak about something important to investors, the legal question can change.

The issue becomes whether what the company said was materially misleading because important information necessary to understand those statements was omitted.

That is essentially what Stewart is asking the court to decide.

The alleged disconnect became public on July 16.

According to the complaint, Bloomberg reported that Google was months behind schedule on Gemini 3.5 Pro and that a change involving the model's training data had produced disappointing results.

Alphabet's Class A shares subsequently fell $16.46, or approximately 4.4%, closing at $354.46. Class C shares dropped $16.39, also approximately 4.4%, to $353.58, according to figures cited in the complaint. Open Class Actions

Those movements matter because securities-fraud plaintiffs generally need more than evidence that a company made an inaccurate statement. They ultimately must establish elements including material misrepresentation or omission, scienter—or wrongful state of mind—and a connection between the alleged deception and investors' financial losses.

That means one of the most important questions in Stewart could eventually become: What did Alphabet's senior leadership actually know about Gemini 3.5 Pro's performance when those public statements were made?

The lawsuit arrives at an interesting moment in the AI industry's evolution.

For decades, investors have scrutinized drug-trial results, manufacturing problems, oil discoveries and product delays for clues about a company's future value.

AI models may now belong on that list.

A disappointing training run inside a technology company can sound highly technical. But if billions of dollars in investment, infrastructure spending and future revenue expectations depend upon increasingly powerful AI models, an internal training problem potentially becomes much more than an engineering issue.

It can become information investors consider when deciding what a company is worth.

That does not mean Stewart will ultimately prove securities fraud.

Alphabet can challenge whether the statements were materially false or misleading, whether executives acted with the required intent, whether investors' losses were actually caused by the alleged disclosures and whether the complaint meets the demanding pleading standards governing federal securities cases.

There isn't even a certified class yet.

Under the federal securities-class-action process, investors seeking appointment as lead plaintiff have 60 days following publication of the required notice to ask the court for that role. The defendants would ordinarily get an opportunity later to seek dismissal before the litigation reaches discovery and class certification.

That makes Friday's filing the beginning—not the conclusion—of the case.

But Stewart v. Alphabet could present a distinctly modern securities-law problem.

As Wall Street increasingly values technology companies according to their ability to produce the next generation of artificial intelligence, what an AI model is doing inside a training cluster may eventually matter almost as much to investors as what a CEO says about it onstage.