AiNews.com
Posts
OpenAI Shares Safety Updates Following Whistleblower Complaints

OpenAI Shares Safety Updates Following Whistleblower Complaints

Alicia Shapiro
July 24, 2024 • Estimated Reading Time: 4 minutes

An illustration showing OpenAI's commitment to safety. The image features elements such as AI models with safety scorecards, a framework for AI preparedness, and a committee reviewing safety protocols. Symbols of transparency, security, and compliance represent the updated whistleblower policies. The scene conveys a focus on AI safety and regulatory compliance.

OpenAI Shares Safety Updates Following Whistleblower Complaints

OpenAI has responded to growing concerns about the safety of generative artificial intelligence models by sharing updates on its safety initiatives. This comes after a seven-page complaint filed by whistleblowers, obtained by The Washington Post, and recent demands from US lawmakers for more transparency.

Safety Prioritization Amid Concerns

On Tuesday, OpenAI took to X (formerly Twitter) to provide an overview of "how we're prioritizing safety in our work." The thread outlined recent initiatives and ongoing projects aimed at enhancing safety in AI development. (You can click on the image below to be taken to X to read the entire thread.)

A screenshot of an X thread from OpenAI, posted on July 23, 2024. The thread discusses how OpenAI is prioritizing safety in its work, highlighting the Preparedness Framework, which helps evaluate and protect against risks posed by powerful AI models. OpenAI states that they won't release a new model if it crosses a "medium" risk threshold. The thread also mentions the development of levels to help categorize and track AI progress, with more details to be shared soon.

Image Source: Screenshot of OpenAI Thread on X

Preparedness Framework

OpenAI highlighted its Preparedness Framework, initially released in beta in December. This framework includes several precautions to ensure the safety of its advanced AI models. It features model scorecards that evaluate various metrics, and a policy of not releasing new models if they exceed a "medium" risk threshold.

AI Progress Levels

The company is actively developing "levels" to help OpenAI and stakeholders categorize and track AI progress. OpenAI plans to share more details about these levels soon, providing stakeholders with a clearer understanding of AI development stages.

Safety and Security Committee

OpenAI also updated the public on its Safety and Security Committee, launched by the board of directors in May. This committee, comprising technical and policy experts, is conducting a review of OpenAI's operations. Upon conclusion, OpenAI will outline further steps to enhance its safety protocols.

Whistleblower Policy and Employee Rights

Addressing concerns from the whistleblower report, OpenAI emphasized its whistleblower policy, which protects employees' rights to make protected disclosures. The company has also revised its employee departure process, removing non-disparagement terms to foster open discussions about the technology.

Response to Lawmaker Demands

These updates were shared a day after US lawmakers demanded that OpenAI provide data on its safety practices in light of the whistleblower report. The report highlighted that OpenAI allegedly prevented staff from alerting authorities about technology risks and made employees waive their federal rights to whistleblower compensation. OpenAI's recent updates aim to address these concerns and enhance transparency.