At a San Francisco playground, a stranger tried to befriend an OpenAI employee who was there with their child. The stranger also regularly showed up with a child, and, in the process of making small talk, asked the OpenAI employee questions about their work. It was no idle curiosity: The stranger was a spy.

That is one of the scenarios employees of the artificial intelligence company have been presented with in attempts to raise their vigilance against attempts to steal corporate secrets—and that was before federal authorities arrested an employee for planting a listening device.

In training sessions, OpenAI has warned employees that outsiders, including competitors and state-sponsored actors, could try to get access to company information. Company leaders have said that could take the form of people offering money, listening in cafes or even trying to seduce OpenAI employees, people familiar with the matter said. Employees at other AI heavyweights, including Anthropic and Google, have received similar warnings.

The companies leading the race to build artificial superintelligence describe the technology as a force that could reshape entire economies, create vast fortunes and dominate future warfare, potentially altering the geopolitical power balance. All that has inevitably made the industry, its employees and their secrets ripe targets for espionage. And while Silicon Valley has been on guard against the threat, there is no such thing as perfect defense.

On Tuesday, a Saudi employee of OpenAI, Abdulrahman Alfozan, was charged with trying to record audio using a device planted behind a TV set in a company conference room. In announcing Alfozan’s arrest, FBI Director Kash Patel said protecting the technological secrets behind AI “is essential to safeguarding our nation’s economic strength and national security, and the FBI views that mission as one of our highest counterintelligence priorities.”Alfozan’s attorney said the allegations were “wholly out of character” and noted his client was charged only with violating a federal recording statute, not stealing trade secrets or acting as a foreign agent.

Corporate security experts say it is likely the threat goes well beyond one hidden recording device. “Catching one guy means there are half a dozen out there,” said Randy Watt, president of leadership consulting firm SRW, which conducts security and threat assessments for companies. He said bad actors may be searching not only for technical or strategic information but for personal secrets that can be used to exploit or blackmail individuals.

The rapid growth of the industry has created abundant opportunities to get inside the leading companies, which have quickly added thousands of employees to their payrolls. OpenAI, Anthropic, Meta and Google all employ large numbers of foreign nationals, with China supplying much of the world’s top AI talent. A number of Trump administration officials have identified employment of Chinese nationals by AI and technology companies as a national security vulnerability, alleging that some share secrets with Beijing.

Employees also frequently job-hop between companies, creating an additional vector for information leakage or misappropriation. OpenAI and Apple have been engaged in litigation over claims of poached employees taking confidential files with them to their new jobs. Elon Musk’s xAI made a similar claim in a lawsuit against an engineer who went to OpenAI.

“AI companies are scaling their workforces at unprecedented speed. The question is whether their ability to manage insider risk is scaling at the same pace,” said Aviv Nahum, the chief executive of Above Security, a startup that uses AI to help companies detect insider risks.

He said the traditional approach to mitigating insider risk relies on verifying someone’s identity and then trusting that what they do afterward is legitimate, which can miss cases where the threat develops over time.

In corporate training sessions, OpenAI instructs employees to avoid wearing anything bearing OpenAI’s logo outside of the office, to use screen protectors that restrict visibility on public transit, and to wipe down whiteboards when they are finished with them, in case kitchen and cleaning staff are compromised, according to people familiar with the matter. Employees are frequently reminded of these precautions, including via Slack, some of the people said.

In some trainings, OpenAI offers additional warnings about scenarios involving “honey pots,” or agents who use the offer of sex or romance to get close to a target, some of the people said.

One story OpenAI tells employees to illustrate the risk: To seduce a U.S. diplomat in Russia, a foreign espionage operation first sent a blond woman to his hotel room with Champagne. When he declined, the spies sent a redheaded woman, and then next a young man. The implication for OpenAI employees was this could happen to them, according to the people.

OpenAI maintains a security team with experience in physical security, insider risk, information security and related disciplines, according to a spokesperson.

Corporate espionage has been going on as long as there has been intellectual property to steal, said Phil Venables, a partner at Ballistic Ventures and investor in companies including Nahum’s, and former chief information security officer for Google Cloud.

In 2022, a former Twitter employee was found guilty of spying for Saudi Arabia and sentenced to 42 months in prison. The man, Ahmad Abouammo, had passed on private user information associated with critics of the kingdom in exchange for hundreds of thousands of dollars, prosecutors said during the trial.

In January, a former Google software engineer was convicted of stealing thousands of pages of confidential information relating to the company’s AI trade secrets and giving the material to China.

According to Venables, the concentration of so much value in a handful of AI companies—OpenAI and Anthropic are each expected to seek valuations of more than $1 trillion in their initial public offerings—makes them bigger targets.

“The resources and money required to train foundation models makes them even more valuable intellectual property,” he said.

Write to Georgia Wells at georgia.wells@wsj.com, Erin Woo at erin.woo@wsj.com and Meghan Bobrowsky at meghan.bobrowsky@wsj.com