OpenAI Fires Three Safety Researchers Over Disappearing Signal Chats
OpenAI has fired three researchers who worked on AI safety. An internal investigation found that they had used Signal, a private messaging app, to talk with a safety expert outside the company. They had turned on a feature that automatically deletes messages.
The firings show growing problems inside OpenAI's safety team. They also show how much pressure employees face and how closely the company controls information about its AI technology. The Wall Street Journal reported that OpenAI's security staff found the conversations during a routine security check. The researchers had been discussing sensitive company information using messages set to disappear.
Changes at OpenAI and Security Problems
The firings happened while OpenAI was making bigger changes to its safety and policy teams. The company was also moving from a business structure that limits profits to one that operates fully for profit. The researchers were suspected of sharing private information about OpenAI's safety tests and what its AI models can do with an outside AI safety organization. Signal is popular among technology workers and journalists because it protects messages with encryption, meaning only the people in a conversation can read them. But the researchers' decision to use automatically disappearing messages raised security concerns inside OpenAI. The company has strict rules that stop employees from sharing confidential information. It also carefully protects its private technology and data, especially the numerical values that determine how its advanced AI models work, the information used to train them, and the methods used to make them behave safely.
Sharing this kind of information with outsiders without permission is considered a serious violation of company rules and trust. The incident has started a debate among AI researchers about how much information companies should keep secret and how much they should share with scientists. People who support stronger AI safety measures say independent experts need to review AI systems to help prevent extremely serious dangers. This is especially important for red-teaming, where experts try to find weaknesses in AI systems, and interpretability research, which tries to understand how AI makes decisions. But if someone leaks the results of safety tests, people with harmful intentions might learn how to get around safety protections or take advantage of weaknesses before they are fixed. On the other hand, OpenAI and competing companies argue that revealing what advanced AI systems can do too early could help rival countries misuse the technology or speed up an AI competition without proper rules.
Problems Inside OpenAI and What Happened Before
This is not the first time OpenAI has had serious disagreements or lost employees from its safety teams. In 2024, the company shut down its Superalignment team. Important members left, saying OpenAI was focusing too much on releasing products quickly and not enough on building a strong culture of safety. The latest firings suggest that OpenAI is becoming stricter about making employees follow its rules as the company grows and looks for more investment money. There is also disagreement between people who want to pause advanced AI development until stronger rules are in place and people who want AI technology to develop and be released quickly. The second approach is often called effective accelerationism. These disagreements have created a tense workplace where protecting company information is extremely important.
The firings appear to have happened as part of a series of steps to enforce security rules. In early October, technology news websites reported that OpenAI had ended its working relationships with several safety researchers. At first, the departures were described as part of normal company restructuring. However, a more detailed Wall Street Journal report published on October 23 explained what reportedly caused the security investigation. It identified the researchers' use of Signal's disappearing messages feature as the main reason for the security concerns. The outside organization suspected of receiving the leaked information has not been publicly named. However, people familiar with the AI industry suggest that it belongs to a group of independent AI safety organizations that regularly criticize the way companies turn advanced AI models into commercial products.
Government Rules and Standards in the AI Industry
The incident also shows a growing difference between how university researchers and private AI companies share information. At universities, researchers normally share early versions of their papers and discuss how AI models behave with other experts. This helps other scientists repeat their experiments, check their results, and build on their discoveries. But at an AI company worth billions of dollars, these conversations are much more tightly controlled because the company wants to protect its advantages over competitors. The fired researchers worked closely on the technical methods used to make large language models follow safety goals. They may have believed that talking to outside experts was an important way to protect people and make sure independent researchers could check OpenAI's claims about safety. OpenAI's leaders, however, see sharing private information without permission as a direct danger to the company's ability to succeed financially and maintain important business partnerships.
Government regulators are also paying more attention to how companies handle sensitive AI research. The US Department of Commerce and government agencies responsible for export controls have strict rules about sending advanced AI technology and model weights to others. Model weights are the numerical values that help determine how an AI system works. The researchers' conversations were probably about making AI systems safer rather than sharing model weights covered by export restrictions. Even so, the incident shows how difficult it can be for technology companies to follow government rules while controlling who can access and share their internal knowledge.
Legal experts who work with technology companies say encrypted messaging apps that automatically delete messages are receiving more attention during company lawsuits and government investigations. Employees often use these apps to protect their privacy. However, automatically deleting work messages can break company rules that require records to be kept for legal compliance and internal checks. The firings send a strong warning to the wider technology industry that casually sharing information about AI development is becoming less acceptable. Instead, companies are introducing very strict security rules, similar to those used to protect sensitive government information. These rules are meant to protect AI technology and research that many people now consider among the most valuable company-owned knowledge in the world.
As OpenAI gets ready to develop its next generation of AI models that can work through complex problems, the firing of three safety researchers over private, encrypted messages shows that the company is tightening its internal security. OpenAI is putting greater importance on protecting its technology and research, even while many scientists are asking AI companies to be more open about their work.